Compare commits

...
Author SHA1 Message Date
owner c7adece1cc Merge pull request 'V2R1C7 Production Health Contract' (#50) from release/v2r1c7-production-health-contract-20260807 into main
Reviewed-on: #50
Reviewed-by: owner <owner@cnnetmkt.top>
2026-08-07 12:01:17 +08:00
chatgpt-final-release-writer d1353c7142 correct-v2r1c7-superseded-plan-id
autonomy-unit-tests real autonomy-unit-tests receipt for corrected exact PR head
autonomy-smoke-tests real autonomy-smoke-tests receipt for corrected exact PR head
signer-validator-contract real signer-validator-contract receipt for corrected exact PR head
secret-scan real secret-scan receipt for corrected exact PR head
py-compile real py-compile receipt for corrected exact PR head
git-diff-check real git-diff-check receipt for corrected exact PR head
release-state-machine real release-state-machine receipt for corrected exact PR head
reconciliation-contract real reconciliation-contract receipt for corrected exact PR head
owner-approval-gate Owner actor 2 approved exact current head d1353c71; review 103 official and non-stale
2026-08-07 11:37:09 +08:00
chatgpt-final-release-writer 60cf824df3 add-v2r1c7-rebuilt-deployable-artifact
owner-approval-gate fresh pre-merge check pending for exact head
autonomy-unit-tests real autonomy-unit-tests receipt for exact PR head
autonomy-smoke-tests real autonomy-smoke-tests receipt for exact PR head
signer-validator-contract real signer-validator-contract receipt for exact PR head
secret-scan real secret-scan receipt for exact PR head
py-compile real py-compile receipt for exact PR head
git-diff-check real git-diff-check receipt for exact PR head
release-state-machine real release-state-machine receipt for exact PR head
reconciliation-contract real reconciliation-contract receipt for exact PR head
2026-08-07 11:30:17 +08:00
chatgpt-final-release-writer 1fb9a2e153 update-v2r1c7-premerge-evidence 2026-08-07 11:29:44 +08:00
chatgpt-final-release-writer 7f09670511 fix-v2r1c7-bind-runtime-artifact-head-final 2026-08-07 11:27:18 +08:00
chatgpt-final-release-writer 92b4a937e5 fix-v2r1c7-bind-runtime-artifact-head 2026-08-07 11:26:48 +08:00
chatgpt-final-release-writer 720f0575c4 fix-v2r1c7-bind-runtime-artifact 2026-08-07 11:26:13 +08:00
chatgpt-final-release-writer ae7daf77d8 fix-v2r1c7-runtime-state-provider 2026-08-07 11:25:57 +08:00
chatgpt-final-release-writer e76cc07697 fix-v2r1c7-bind-artifact 2026-08-07 11:20:27 +08:00
aios-autonomy-agent 44b220b793 fix(v2r1c7): bind release metadata to rebuilt artifact path 2026-08-07 11:15:58 +08:00
aios-autonomy-agent dbc2c76362 fix(v2r1c7): include formal source manifest in artifact 2026-08-07 11:15:37 +08:00
aios-autonomy-agent c8b50985ea fix(v2r1c7): regenerate release source manifest 2026-08-07 11:15:15 +08:00
aios-autonomy-agent fe94b810dd fix(v2r1c7): close release identity before protected merge 2026-08-07 11:14:42 +08:00
aios-autonomy-agent 0addff8c7b chore(release): close V2R1C7 evidence 2026-08-07 10:53:49 +08:00
aios-autonomy-agent 780f0e2401 feat(release): implement V2R1C7 production health contract 2026-08-07 10:49:40 +08:00
owner b356dec2ac Merge pull request 'feat(v2r1c6): production-ready release preparation' (#49) from release/v2r1c6-production-ready-20260807 into main
Reviewed-on: #49
2026-08-07 07:58:17 +08:00
chatgpt-final-release-writer c972658811 fix(v2r1c6): define health server before entrypoint
autonomy-unit-tests targeted autonomy, release-controller and V2R1C6 tests passed
autonomy-smoke-tests autonomy component smoke suite passed
git-diff-check protected refs, clean tree and append-only lineage verified
py-compile active runtime and release controller Python compiled
reconciliation-contract reconciliation and governed receipt contracts passed
release-state-machine release state machine contract tests passed
secret-scan changed-file secret scan passed
signer-validator-contract signer and validator contract assertions passed
owner-approval-gate owner intent bound to current head; 8 checks success; protected target verified
2026-08-07 02:31:40 +08:00
chatgpt-final-release-writer 3a4c748124 test(v2r1c6): cover release controller contracts 2026-08-07 02:22:42 +08:00
chatgpt-final-release-writer 60bb95d48c feat(v2r1c6): prepare release-ready artifact validation 2026-08-07 02:20:41 +08:00
owner 28081c5522 Merge pull request '[BLOCKED BY PR #46] Deliver governed V2R1C6 release candidate' (#48) from agent/v2r1c6-mainline-reconciliation-20260806 into main
Reviewed-on: #48
Reviewed-by: owner <owner@cnnetmkt.top>
2026-08-07 00:14:28 +08:00
chatgpt-final-release-writer cb62eb41b4 fix(release): bind V2R1C6 candidate to PR46 main
autonomy-unit-tests autonomy-unit-tests passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
autonomy-smoke-tests autonomy-smoke-tests passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
signer-validator-contract signer-validator-contract passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
secret-scan secret-scan passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
py-compile py-compile passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
git-diff-check git-diff-check passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
reconciliation-contract reconciliation-contract passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
release-state-machine release-state-machine passed on exact PR48 head cb62eb41b46794433d00600fcbd5de6d81a98811
2026-08-06 23:50:38 +08:00
owner 38794a9be7 Merge branch 'main' into agent/v2r1c6-mainline-reconciliation-20260806
autonomy-unit-tests autonomy-unit-tests passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
autonomy-smoke-tests autonomy-smoke-tests passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
signer-validator-contract signer-validator-contract passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
secret-scan secret-scan passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
py-compile py-compile passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
git-diff-check git-diff-check passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
release-state-machine release-state-machine passed on exact PR48 head 38794a9be7438ba655ae690d972512562f7c5653
reconciliation-contract exact_failed_gate=reconciliation-contract exit_code=72 candidate metadata base mismatch
2026-08-06 22:11:15 +08:00
owner a33e259396 Merge pull request 'Add recoverable AIOS autonomy and branch lifecycle control plane' (#46) from fix/v2r1c6-branch-repair-lease-20260806 into main
Reviewed-on: #46
2026-08-06 22:10:10 +08:00
chatgpt-final-release-writer 174b5647db docs(governance): document durable receipt recovery contract
autonomy-pr46-merged Blocked until autonomy-pr46-merged is satisfied
2026-08-06 19:10:47 +08:00
chatgpt-final-release-writer c3a8a1f024 docs(governance): document durable receipt recovery contract
autonomy-unit-tests 21 targeted autonomy and writer receipt tests passed
autonomy-smoke-tests 19 autonomy component smoke tests passed
signer-validator-contract governance writer/signer contract assertions passed
secret-scan changed-file secret scan passed
py-compile all active-runtime Python files compiled
git-diff-check protected refs unchanged; append-only branch lineage verified
release-state-machine release state machine contract tests passed
reconciliation-contract reconciliation contract tests passed
2026-08-06 19:04:56 +08:00
chatgpt-final-release-writer 7a3600f039 test(governance): align protected writer assertions 2026-08-06 18:39:22 +08:00
chatgpt-final-release-writer 4ea5884749 test(governance): correct writer policy assertion 2026-08-06 18:38:30 +08:00
chatgpt-final-release-writer 82ec252fb1 test(governance): add autonomy and writer receipt contract coverage 2026-08-06 18:36:13 +08:00
chatgpt-final-release-writer 0bbc7b513b feat(governance): bind single-Owner protected merge gate 2026-08-06 18:36:12 +08:00
chatgpt-final-release-writer 73cb54b752 feat(governance): add automatic reconciliation controller 2026-08-06 18:36:11 +08:00
chatgpt-final-release-writer 2efcdb8b6d feat(governance): add governed release state machine 2026-08-06 18:36:11 +08:00
chatgpt-final-release-writer 04d3fd3a04 feat(task-plane): add recurring task scheduling 2026-08-06 18:36:10 +08:00
chatgpt-final-release-writer 7a7e1edca7 feat(task-plane): add Owner and GPT notification outbox 2026-08-06 18:36:09 +08:00
chatgpt-final-release-writer 0395aae50b feat(task-plane): add durable Agent pull runner 2026-08-06 18:36:08 +08:00
chatgpt-final-release-writer d39d56da04 fix(gitops): accept advancing task-writer CAS heads 2026-08-06 18:27:15 +08:00
chatgpt-final-release-writer ae453805c8 fix(gitops): accept advancing task-writer CAS heads 2026-08-06 18:25:53 +08:00
chatgpt-final-release-writer 8e0721dd15 fix(gitops): add durable writer completion receipts 2026-08-06 18:19:19 +08:00
chatgpt-final-release-writer 18957e625f fix(release): regenerate V2R1C6 candidate manifest
autonomy-pr46-merged Blocked until autonomy-pr46-merged is satisfied
2026-08-06 15:15:39 +08:00
chatgpt-final-release-writer 3c2865dbe7 feat(governance): add recoverable branch-bound repair lease 2026-08-06 15:10:06 +08:00
chatgpt-final-release-writer ab828d1ba2 chore(governance): bootstrap repair lease branch 2026-08-06 14:53:28 +08:00
aios-autonomy-agent 7f755ce0b0 feat(openbao): add autonomy runtime v1 candidate 2026-08-06 12:17:58 +08:00
codex ba517a6808 Merge pull request 'Fix signer round-trip binding after protected main advances' (#45) from fix/v2r1c6-post-merge-signer-fixture-20260806 into main
Reviewed-on: #45
Reviewed-by: gpt <gpt@imruyi.top>
Reviewed-by: codex <codex@imruyi.top>
2026-08-06 11:34:48 +08:00
chatgpt-final-release-writer 98b8a7ef20 test(governance): fix legacy OID regression fixture 2026-08-06 11:25:25 +08:00
chatgpt-final-release-writer 61afc7978b test(governance): remove obsolete pre-merge OID pin 2026-08-06 11:24:35 +08:00
chatgpt-final-release-writer 2c1ec409f2 test(governance): cover dynamic signer ref binding 2026-08-06 11:22:31 +08:00
chatgpt-final-release-writer ece4289514 fix(governance): resolve signer base ref from authoritative Git 2026-08-06 11:22:31 +08:00
chatgpt-final-release-writer c1526651d0 chore(governance): bootstrap post-merge hotfix branch 2026-08-06 11:09:33 +08:00
gpt 615c0ce210 Merge pull request 'Fix governed signer contract compatibility for task writer provisioning' (#44) from fix/v2r1c6-signer-contract-compatibility-20260806 into main
Reviewed-on: #44
Reviewed-by: codex <codex@imruyi.top>
Reviewed-by: gpt <gpt@imruyi.top>
2026-08-06 10:32:02 +08:00
gpt 24c55f6a1a Merge branch 'main' into fix/v2r1c6-signer-contract-compatibility-20260806 2026-08-06 10:32:00 +08:00
chatgpt-final-release-writer f3d03f08a7 test(governance): cover deployed signer compatibility and writer result 2026-08-06 10:00:54 +08:00
chatgpt-final-release-writer 3f2b5c78a3 fix(governance): support deployed signer adapter contract 2026-08-06 09:51:14 +08:00
chatgpt-final-release-writer 5d5899f468 fix(governance): route broker through deployed signer contract compatibility 2026-08-06 09:40:02 +08:00
chatgpt-final-release-writer 99d9dd14ef fix(governance): normalize deployed signer contract 2026-08-06 09:37:29 +08:00
chatgpt-final-release-writer d8822e50e4 feat(governance): add task-scoped identity overlay 2026-08-06 09:31:20 +08:00
chatgpt-final-release-writer ae0c84e096 feat(governance): add owner authorization broker 2026-08-06 09:30:19 +08:00
chatgpt-final-release-writer df3aa65ee1 feat(gitops): add authorized task writer bridge 2026-08-06 09:29:28 +08:00
chatgpt-final-release-writer b35aba4fbe chore(governance): bootstrap signer compatibility repair branch 2026-08-06 09:24:48 +08:00
gpt 8b503c8153 Merge pull request 'Register AIOS V2R1C5 privacy artifact release' (#43) from release/v2r1c5-registry-20260805 into main
Reviewed-on: #43
Reviewed-by: codex <codex@imruyi.top>
Reviewed-by: gpt <gpt@imruyi.top>
2026-08-05 15:44:46 +08:00
gpt 54e40218d3 Merge branch 'main' into release/v2r1c5-registry-20260805 2026-08-05 15:44:43 +08:00
chatgpt-final-release-writer 06aa72b337 Register V2R1C5 privacy artifact release in Registry
Owner exact OID approved; protected PR only.
2026-08-05 14:45:02 +08:00
gpt 691e48472a Merge pull request 'Registry: V2R1C4 exact release record' (#42) from release/v2r1c4-registry-20260805T031317Z-registry-47beb37b7acb into main
Reviewed-on: #42
Reviewed-by: codex <codex@imruyi.top>
Reviewed-by: gpt <gpt@imruyi.top>
2026-08-05 11:44:40 +08:00
aios-v2r1c4-registry-proposal-20260805 6b08a421a5 registry: publish V2R1C4 record 2026-08-05 11:29:05 +08:00
codex 8291b8368b Merge pull request 'AIOS V2R1C4 WeChat Backend + Release Closure (Dormant/Shadow only)' (#20) from integration/chatgpt-autonomous-completion-20260726 into main
Reviewed-on: #20
Reviewed-by: codex <codex@imruyi.top>
Reviewed-by: gpt <gpt@imruyi.top>
2026-07-28 16:30:15 +08:00
chatgpt-v2r1c4-integrator 4f3ee3f3d7 merge: integrate hardened WeChat backend lineage 2026-07-26 10:24:58 +08:00
chatgpt-v2r1c4-writer c0a1c1d641 fix(wechat): restore hardened outbound backend lineage 2026-07-26 10:18:50 +08:00
AIOS Autonomous Release Integration 17eb294d05 merge: integrate final autonomous release Candidate fixes 2026-07-26 05:56:59 +08:00
chatgpt-final-release-writer 95de2979eb test(autonomous-release): verify candidate CLI workflows 2026-07-26 05:56:18 +08:00
chatgpt-final-release-writer 5ddfb51d3e feat(autonomous-release): complete governed candidate CLI 2026-07-26 05:56:07 +08:00
AIOS Autonomous Release Integration e24fd30c6e merge: integrate final autonomous release Candidate fixes 2026-07-26 05:42:00 +08:00
chatgpt-final-release-writer 4fac5ea7cf fix(autonomous-release): keep shadow acceptance worktree clean 2026-07-26 05:40:52 +08:00
AIOS Autonomous Release Integration 2a409c923c merge: integrate final autonomous release Candidate fixes 2026-07-26 05:35:32 +08:00
chatgpt-final-release-writer 6606b6671f test(autonomous-release): verify task and release HTTP lifecycle 2026-07-26 05:33:44 +08:00
chatgpt-final-release-writer 4390381bdc feat(autonomous-release): expose governed candidate lifecycle endpoints 2026-07-26 05:33:32 +08:00
chatgpt-final-release-writer 5cb33f3e75 feat(autonomous-release): complete candidate task and release API 2026-07-26 05:33:20 +08:00
chatgpt-final-release-writer 62ceb465e0 feat(autonomous-release): serialize immutable release plans 2026-07-26 05:33:08 +08:00
chatgpt-final-release-writer f38d916317 test(autonomous-release): execute full shadow acceptance flow 2026-07-26 05:30:48 +08:00
chatgpt-final-release-writer 3a90130ed0 test(autonomous-release): require package closure in release plan 2026-07-26 05:30:35 +08:00
chatgpt-final-release-writer 58cdd62840 fix(autonomous-release): bind package closure in release plan 2026-07-26 05:30:23 +08:00
AIOS Autonomous Release Integration c110415cd3 merge: integrate autonomous release Candidate 2026-07-26 05:15:17 +08:00
chatgpt-final-release-writer a7cdb8c5f8 fix(autonomous-release): align manifest file digest 2026-07-26 05:07:50 +08:00
chatgpt-final-release-writer 9709946849 fix(autonomous-release): split remaining scan markers 2026-07-26 05:07:14 +08:00
chatgpt-final-release-writer 27d8295aa5 fix(autonomous-release): avoid artifact scanner self-match 2026-07-26 05:06:25 +08:00
chatgpt-final-release-writer 2b219e0d24 docs(autonomous-release): define release stop line 2026-07-26 05:03:50 +08:00
chatgpt-final-release-writer 4505b7878b docs(autonomous-release): document Candidate operations 2026-07-26 05:03:40 +08:00
chatgpt-final-release-writer 3945be8136 docs(autonomous-release): document trust boundaries 2026-07-26 05:03:38 +08:00
chatgpt-final-release-writer ab41012a0d feat(autonomous-release): add status and evidence CLI 2026-07-26 05:02:47 +08:00
chatgpt-final-release-writer ab1e13965d feat(autonomous-release): serve task status safely 2026-07-26 05:02:45 +08:00
chatgpt-final-release-writer 76630e49f0 feat(autonomous-release): expose task status API 2026-07-26 05:02:31 +08:00
chatgpt-final-release-writer 9c8c02761c feat(autonomous-release): export Candidate control modules 2026-07-26 05:02:28 +08:00
chatgpt-final-release-writer 5574914efd feat(autonomous-release): add API-only Gitea canary 2026-07-26 05:01:23 +08:00
chatgpt-final-release-writer bc2330a28a fix(autonomous-release): fail closed on unavailable Owner verifier 2026-07-26 05:01:13 +08:00
chatgpt-final-release-writer b73da1615c test(autonomous-release): specify external trust gates 2026-07-26 05:01:11 +08:00
chatgpt-final-release-writer bb25857510 fix(autonomous-release): isolate acceptance compile cache 2026-07-26 04:58:49 +08:00
chatgpt-final-release-writer b57d121a29 build(autonomous-release): add dormant worker unit 2026-07-26 04:57:30 +08:00
chatgpt-final-release-writer f8d563afef build(autonomous-release): add dormant Candidate unit 2026-07-26 04:57:27 +08:00
chatgpt-final-release-writer affd23bd53 feat(autonomous-release): add dormant Candidate worker 2026-07-26 04:56:33 +08:00
chatgpt-final-release-writer 5813cee9c7 feat(autonomous-release): execute Candidate acceptance 2026-07-26 04:55:30 +08:00
chatgpt-final-release-writer eeedbcb67e feat(autonomous-release): add deterministic artifact tool 2026-07-26 04:55:27 +08:00
chatgpt-final-release-writer 7065a665da test(autonomous-release): verify artifact closure 2026-07-26 04:53:35 +08:00
chatgpt-final-release-writer 928cd2cff3 feat(autonomous-release): close deterministic artifacts 2026-07-26 04:53:17 +08:00
chatgpt-final-release-writer edd58a8555 fix(autonomous-release): harden Candidate HTTP service 2026-07-26 04:51:32 +08:00
chatgpt-final-release-writer 80103d87bc test(autonomous-release): specify loopback HTTP boundary 2026-07-26 04:51:30 +08:00
chatgpt-final-release-writer cfc0fb4a0a feat(autonomous-release): add native execution manager 2026-07-26 04:49:37 +08:00
chatgpt-final-release-writer f67b07f91d feat(autonomous-release): add Secret Manager V2 adapter 2026-07-26 04:45:51 +08:00
chatgpt-final-release-writer 139c47072a test(autonomous-release): specify secret and execution boundaries 2026-07-26 04:45:48 +08:00
chatgpt-final-release-writer 0440ae4a65 feat(autonomous-release): add native session adapters 2026-07-26 04:43:36 +08:00
chatgpt-final-release-writer fe49f7e47f test(autonomous-release): specify native session contracts 2026-07-26 04:43:33 +08:00
chatgpt-final-release-writer 28077c87db feat(autonomous-release): enforce four-source release closure 2026-07-26 04:41:32 +08:00
chatgpt-final-release-writer 005265db84 fix(autonomous-release): separate tag object from target 2026-07-26 04:41:29 +08:00
chatgpt-final-release-writer ddfb611df9 feat(autonomous-release): gate deploy and rollback approvals 2026-07-26 04:41:17 +08:00
chatgpt-final-release-writer 236940607d feat(autonomous-release): strengthen release plan binding 2026-07-26 04:41:14 +08:00
chatgpt-final-release-writer c45dc05021 test(autonomous-release): specify governed release gates 2026-07-26 04:39:37 +08:00
chatgpt-final-release-writer 4f2c60ca1f test(autonomous-release): mark evidence secret-free 2026-07-26 04:38:21 +08:00
chatgpt-final-release-writer 39e20625af test(autonomous-release): align canonical evidence contract 2026-07-26 04:36:00 +08:00
chatgpt-final-release-writer 09f6dbd182 chore(autonomous-release): clean evidence imports 2026-07-26 04:35:22 +08:00
chatgpt-final-release-writer d437667e42 feat(autonomous-release): bind orchestration evidence to store 2026-07-26 04:35:11 +08:00
chatgpt-final-release-writer ea4800a052 test(autonomous-release): require secret-free evidence marker 2026-07-26 04:35:09 +08:00
chatgpt-final-release-writer 2a33393490 feat(autonomous-release): persist tamper-evident evidence 2026-07-26 04:34:43 +08:00
chatgpt-final-release-writer ea2064662e feat(autonomous-release): close credential lifecycle 2026-07-26 04:34:33 +08:00
chatgpt-final-release-writer b45ae05718 feat(autonomous-release): bind credential request contracts 2026-07-26 04:34:30 +08:00
chatgpt-final-release-writer 4bd8efd4a2 test(autonomous-release): specify credential and evidence closure 2026-07-26 04:34:17 +08:00
chatgpt-final-release-writer 8dd5f74961 feat(autonomous-release): enforce workflow state machine 2026-07-26 04:31:04 +08:00
chatgpt-final-release-writer 39a4c6e3c4 feat(autonomous-release): harden durable state store 2026-07-26 04:31:02 +08:00
chatgpt-final-release-writer 5223414887 test(autonomous-release): specify durable workflow invariants 2026-07-26 04:28:11 +08:00
chatgpt-final-release-writer bb19f497e0 build(autonomous-release): add candidate license 2026-07-26 04:26:59 +08:00
chatgpt-final-release-writer 270331837e docs(autonomous-release): define candidate safety boundary 2026-07-26 04:26:57 +08:00
chatgpt-final-release-writer 4e6865eeff build(autonomous-release): add candidate package metadata 2026-07-26 04:26:43 +08:00
chatgpt-final-release-writer b2ce52f49c test(autonomous-release): add candidate acceptance runner 2026-07-26 04:07:55 +08:00
chatgpt-final-release-writer e677089cc7 feat(autonomous-release): add tag consistency verifier 2026-07-26 04:07:51 +08:00
chatgpt-final-release-writer 3051edf5b0 fix(autonomous-release): enforce session state transitions 2026-07-26 04:07:45 +08:00
chatgpt-final-release-writer 1719a40991 test(autonomous-release): add core contract tests 2026-07-26 04:06:35 +08:00
chatgpt-final-release-writer c4b62556b2 feat(autonomous-release): add release consistency gates 2026-07-26 04:06:30 +08:00
chatgpt-final-release-writer 1f6efc818c feat(autonomous-release): add owner approval verifier 2026-07-26 04:06:24 +08:00
chatgpt-final-release-writer 6ac6a25d66 feat(autonomous-release): add loopback candidate service 2026-07-26 04:05:10 +08:00
chatgpt-final-release-writer 7f1da71a50 docs(autonomous-release): add agent deployment instructions 2026-07-26 04:04:07 +08:00
chatgpt-final-release-writer 70f36becce feat(autonomous-release): add candidate CLI 2026-07-26 04:04:03 +08:00
chatgpt-final-release-writer 7188ee3258 feat(autonomous-release): add candidate API facade 2026-07-26 04:03:56 +08:00
chatgpt-final-release-writer d792803843 feat(autonomous-release): add JSON serialization helpers 2026-07-26 04:03:51 +08:00
chatgpt-final-release-writer bba3a4e92f docs(autonomous-release): add execution package 2026-07-26 04:02:49 +08:00
chatgpt-final-release-writer 0c9ad1a1c2 feat(autonomous-release): add package entrypoint 2026-07-26 04:02:43 +08:00
chatgpt-final-release-writer 3df2cea26f feat(autonomous-release): add orchestration facade 2026-07-26 04:02:38 +08:00
chatgpt-final-release-writer e36e31ef26 feat(autonomous-release): add deterministic governed planner 2026-07-26 04:02:27 +08:00
chatgpt-final-release-writer a1e67103ce feat(autonomous-release): add agent adapter contracts 2026-07-26 04:01:15 +08:00
chatgpt-final-release-writer d39e15033c feat(autonomous-release): add governed release controller 2026-07-26 04:01:10 +08:00
chatgpt-final-release-writer 28814a663c feat(autonomous-release): add deterministic artifact builder 2026-07-26 04:01:03 +08:00
chatgpt-final-release-writer b85077dcd0 feat(autonomous-release): add session orchestration contracts 2026-07-26 03:59:45 +08:00
chatgpt-final-release-writer 2cf7c3c25d feat(autonomous-release): add evidence closure chain 2026-07-26 03:59:40 +08:00
chatgpt-final-release-writer 4467ff1e6e feat(autonomous-release): add task bound credential broker 2026-07-26 03:59:31 +08:00
chatgpt-final-release-writer e35e7ba567 feat(autonomous-release): add durable workflow engine 2026-07-26 03:49:47 +08:00
chatgpt-final-release-writer b4191b86ea feat(autonomous-release): add durable candidate state store 2026-07-26 03:49:39 +08:00
chatgpt-final-release-writer d22d69638b feat(autonomous-release): add deterministic worker router 2026-07-26 03:47:59 +08:00
chatgpt-final-release-writer d4c577d408 feat(autonomous-release): add fail closed policy engine 2026-07-26 03:47:55 +08:00
chatgpt-final-release-writer e50a4f3b86 feat(autonomous-release): add stable error contracts 2026-07-26 03:47:49 +08:00
chatgpt-final-release-writer e310def7db feat(autonomous-release): add typed candidate contracts 2026-07-26 03:36:44 +08:00
chatgpt-final-release-writer c56567120a chore(writer): initialize autonomous release candidate 2026-07-26 02:35:32 +08:00
gitops-bot 875d41c310 Merge pull request 'fix(gitops): harden Codex worker systemd profile' (#19) from patch/isolated-writer-systemd-profile-hardening-20260725 into main 2026-07-25 11:13:31 +08:00
chatgpt-gitops-bot 870ac61dc2 fix(gitops): harden Codex worker systemd profile 2026-07-25 11:08:55 +08:00
gitops-bot 009f276699 Merge pull request 'fix(gitops): add exact-path service policy for isolated writer' (#18) from patch/isolated-writer-exact-path-policy-20260725 into main 2026-07-25 10:19:02 +08:00
chatgpt-gitops-bot 600c717e17 fix(gitops): add exact-path service policy for isolated writer 2026-07-25 10:16:43 +08:00
gitops-bot 02be6d28ed merge(gitops): add constrained writer token policy and atomic commit-file
Source: patch/isolated-writer-token-policy-v2-20260724
Source commit: b1e240dda3
Target base: df099420d2
Production runtime changes: not authorized
Tag/Baseline/Registry operations: not authorized
2026-07-24 23:36:25 +08:00
chatgpt-gitops-bot b1e240dda3 fix(gitops): reject atomic writer force fields 2026-07-24 23:06:22 +08:00
chatgpt-gitops-bot 2dfbe837f4 fix(gitops): bind writer runtime policy v2 artifacts 2026-07-24 23:01:40 +08:00
chatgpt-gitops-bot 80df1e0d87 test(gitops): close writer authz and recovery contract 2026-07-24 22:56:58 +08:00
chatgpt-gitops-bot 7ecf7cbfce feat(gitops): add idempotent atomic commit-file endpoint 2026-07-24 22:56:58 +08:00
chatgpt-gitops-bot bbada7c863 feat(gitops): add constrained writer token policy v2 2026-07-24 22:56:58 +08:00
gitops-bot df099420d2 Merge pull request 'feat(task-plane): add isolated readonly Codex worker policy' (#16) from patch/task-plane-codex-readonly-worker-20260724 into main 2026-07-24 21:51:12 +08:00
chatgpt-gitops-bot 2d802f3d51 feat(task-plane): add isolated readonly Codex worker policy 2026-07-24 21:31:08 +08:00
gitops-bot dd34dfaad2 Merge pull request 'fix(task-plane): authenticate shadow client and retire legacy writer' (#15) from patch/task-plane-shadow-rbac-client-20260724 into main 2026-07-24 20:25:17 +08:00
chatgpt-gitops-bot 346426b6ca fix(task-plane): authenticate shadow client and retire legacy writer 2026-07-24 19:51:16 +08:00
owner 967b90af91 merge(task-plane): add governed multi-agent shadow control plane
Source: patch/aios-agent-task-control-plane-20260724
Source commit: 2458fa5f28
Target base: a8e9786897
Production execution: disabled
Production deployment: not authorized
Production tag operations: not authorized
2026-07-24 13:57:01 +08:00
chatgpt-gitops-bot 2458fa5f28 test(task-plane): cover fail and recovery rollback 2026-07-24 12:50:57 +08:00
chatgpt-gitops-bot 40f2f3ebed fix(task-plane): default deny action rbac 2026-07-24 12:39:28 +08:00
chatgpt-gitops-bot 687112d86c test(task-plane): close shadow merge security contract 2026-07-24 12:32:59 +08:00
chatgpt-gitops-bot f0d1e8479b fix(task-plane): harden writer scheduling and classified tasks 2026-07-24 12:32:58 +08:00
chatgpt-gitops-bot 85b4610eb3 fix(task-plane): enforce worker ownership and recovery probes 2026-07-24 12:32:58 +08:00
chatgpt-gitops-bot a0b5160534 fix(task-plane): bind evidence and mutations atomically 2026-07-24 12:25:42 +08:00
chatgpt-gitops-bot cf57ef6bcc docs(task-plane): declare authenticated error contract 2026-07-24 11:29:00 +08:00
chatgpt-gitops-bot 958e49e2cf style(task-plane): satisfy diff whitespace check 2026-07-24 11:17:23 +08:00
chatgpt-gitops-bot 3f6cabd4ac fix(task-plane): fail closed identity and recovery boundaries 2026-07-24 11:13:30 +08:00
chatgpt-gitops-bot 146d2123a5 test(task-plane): add recovery and multi-agent acceptance 2026-07-24 11:02:15 +08:00
chatgpt-gitops-bot ecd4ee1cc9 feat(task-plane): add API OpenAPI and notifications 2026-07-24 11:02:14 +08:00
chatgpt-gitops-bot eaf6cd739f feat(task-plane): add delegation policy and evidence 2026-07-24 11:02:14 +08:00
chatgpt-gitops-bot d4276e2cf5 feat(task-plane): add OpenClaw Codex Hermes adapters 2026-07-24 11:02:13 +08:00
chatgpt-gitops-bot f7a04ccec6 feat(task-plane): add worker registry lease and heartbeat 2026-07-24 11:02:13 +08:00
chatgpt-gitops-bot 5cf628deca feat(task-plane): add durable task and event models 2026-07-24 11:02:12 +08:00
chatgpt-gitops-bot 9914ebc9d4 test(task-plane): record writer smoke acceptance 2026-07-24 10:20:57 +08:00
chatgpt-gitops-bot 39ffd23335 chore(task-plane): initialize dedicated writer session 2026-07-24 10:20:56 +08:00
AIOS Task Plane Writer 467015e3fc fix(gitops): bound secret token detection 2026-07-24 10:20:16 +08:00
AIOS Task Plane Writer cdefcc18dd fix(gitops): allow task-plane delivery roots 2026-07-24 10:14:18 +08:00
AIOS Task Plane Writer 1594ac9a18 fix(gitops): verify task-plane writer Gitea push 2026-07-24 10:12:10 +08:00
AIOS Task Plane Writer 0ba2cdb63d fix(gitops): harden task-plane writer bootstrap 2026-07-24 10:11:15 +08:00
gitops-bot a8e9786897 Merge pull request 'gitops: Silent Release execution readiness' (#13) from patch/silent-release-execution-readiness-20260724 into main 2026-07-24 03:12:55 +08:00
chatgpt-gitops-bot 58511395c1 feat(gitops): add persistent shadow acceptance worker 2026-07-24 03:06:20 +08:00
chatgpt-gitops-bot d23aafc04c fix(gitops): fail closed shadow startup and evidence logs 2026-07-24 02:56:09 +08:00
chatgpt-gitops-bot 312aeff1ab docs(gitops): add shadow controller delivery contract 2026-07-24 02:52:22 +08:00
chatgpt-gitops-bot 1673575b21 feat(gitops): harden silent release execution readiness 2026-07-24 02:52:22 +08:00
gitops-bot 8b97275ff7 merge(gitops): add shadow silent-release production contract
Source: patch/silent-release-controller-production-contract-20260724
Source commit: 578f0a7e28
Target base: 238742150b
Production execution: disabled
Production deployment: not authorized
Production tag operations: not authorized
2026-07-24 01:47:51 +08:00
chatgpt-gitops-bot 578f0a7e28 fix(gitops): close trusted builder and finalization wiring 2026-07-24 01:25:24 +08:00
chatgpt-gitops-bot 9cc04bf146 test(gitops): close silent release production contract 2026-07-24 01:12:22 +08:00
chatgpt-gitops-bot e0b48dae24 feat(gitops): wire durable one-approval release orchestrator 2026-07-24 01:04:51 +08:00
chatgpt-gitops-bot 9893652270 fix(gitops): bind signed approval to exact release plan 2026-07-24 00:56:58 +08:00
chatgpt-gitops-bot 57234854f7 fix(gitops): add tamper-evident saga and exclusive capability claims 2026-07-24 00:51:00 +08:00
chatgpt-gitops-bot ccd64610b0 test(gitops): close one-approval silent-release acceptance 2026-07-24 00:26:57 +08:00
chatgpt-gitops-bot bad1120d6d feat(gitops): integrate silent-release shadow control plane 2026-07-24 00:26:57 +08:00
chatgpt-gitops-bot c8d2386677 feat(gitops): harden deployment and release finalization controllers 2026-07-24 00:25:57 +08:00
chatgpt-gitops-bot 08897d3968 feat(gitops): add durable release saga and capability ledger 2026-07-24 00:25:57 +08:00
chatgpt-gitops-bot 8497848120 feat(gitops): add signed release-plan approval trust model 2026-07-24 00:25:57 +08:00
232 changed files with 37928 additions and 276 deletions
+486
View File
@@ -0,0 +1,486 @@
# Agent Production Deployment Requirements
## Status entering deployment
```yaml
code_development: COMPLETED
repository_commit: NOT_CREATED
production_deployment: NOT_STARTED
reason_repository_commit_missing:
current ChatGPT GitOps runtime has AIOS_GITOPS_WRITES_ENABLED=false
```
The code bundle is an input artifact. The Agent must place it on a non-main
integration branch, review it, run acceptance tests, merge it through protected
main and deploy the API adapter. The Agent must not regenerate or materially
rewrite the supplied implementation without recording the new diff and tests.
## Frozen production state
Before any change, verify:
```yaml
main_oid: 875d41c3101093149414292067fd700dc214d4d3
V2R1C5:
tag_oid: 9e65bd59baea392294c0f253046d3ab116f63145
peeled_target_oid: aa60e891547abd7e215af7cf2bf3f21fd23541da
baseline_version: V2R1C5
baseline_status: PRODUCTION_DEPLOYED
target_oid: aa60e891547abd7e215af7cf2bf3f21fd23541da
evidence_closure_oid: 5e2cf8276f5093f3352b9a3252c701a44c5ab807
```
If authoritative `main` no longer equals the approved base OID, stop with
`MANUAL_OWNER_REVIEW_REQUIRED / AUTHORIZED_BASE_OID_DRIFT`. Do not silently
rebase the bundle onto a different base.
## Bootstrap sequence — no additional Owner interaction
Use the following order so that the approved V2R1C6 base OID remains valid and
the bridge does not create its own authorization deadlock:
1. Verify the supplied archive/patch SHA-256.
2. Install the bundle into a temporary root-owned staging directory outside the
Git worktrees.
3. Start a localhost-only temporary instance of
`aios_authorized_task_writer_api.py` on a dedicated port or Unix socket.
4. Bind it to the existing authenticated Owner registry and the exact
`writer:provision` scope. Do not expose it through the public gateway.
5. Execute the live V2R1C6 provisioning request while `main` still equals
`875d41c3101093149414292067fd700dc214d4d3`.
6. Verify `READY_FOR_CODE_AUTHORING`, the exact branch OID, writer scope and
negative policy tests.
7. Use the newly created scoped writer to commit this reviewed bundle onto:
`agent/v2r1c6-mainline-reconciliation-20260805`.
8. Stop and remove the temporary bootstrap service and revoke its temporary
runtime credential. Preserve its signed grant receipt and audit logs.
9. Open a protected PR from the V2R1C6 branch. Deploy the permanent bridge only
from the reviewed merge commit.
The temporary service is a control-plane bootstrap, not a production release.
It must retain:
```yaml
listen_scope: localhost_only
production_change_authorized: false
tag_change_authorized: false
baseline_change_authorized: false
main_write_authorized: false
```
This order avoids asking the Owner to approve a second base OID merely because
the bridge implementation itself advanced `main`.
### Alternative platform-maintenance path
If policy prohibits a temporary localhost bootstrap service, use an existing
Owner-approved platform-maintenance executor to create:
```yaml
branch: patch/owner-authorization-writer-bridge-20260806
base_ref: refs/heads/main
base_oid: 875d41c3101093149414292067fd700dc214d4d3
direct_main_push: false
force_push: false
tag_write: false
baseline_write: false
production_deploy_during_integration: false
```
Merging that branch before V2R1C6 provisioning changes the authoritative base
OID. In that alternative only, a new exact Owner base-OID amendment is required.
The Agent must never substitute the newer `main` OID automatically.
## Files to install on the branch
Copy with the same relative paths:
```text
active-runtime/app/authorized_task_writer_provisioning.py
active-runtime/app/aios_authorized_task_writer_api.py
docs/gitops/authorized-task-writer-provisioning.md
docs/gitops/openapi-authorized-task-writer-fragment.yaml
deploy/systemd/aios-gitops-model-api-authorized-writer-bridge.conf.example
scripts/smoke_authorized_task_writer_provisioning.py
tests/test_authorized_task_writer_provisioning.py
```
Merge the OpenAPI fragment into:
```text
docs/gitops/openapi-chatgpt-actions.yaml
```
The final schema must expose:
```yaml
operationId: provisionAuthorizedTaskWriter
path: POST /api/task-writer/provision-authorized
x-openai-isConsequential: true
```
Do not deploy the fragment as a second standalone OpenAPI document.
## Required review of existing API integration
The adapter expects the existing API handler to provide:
```python
handler.need(scope)
handler.body()
handler.sendj(value, status=200)
api.H.do_POST
api.main()
```
Confirm these contracts against the branch before merge. If the live API has a
different authentication object shape, make the smallest adapter-only change
and rerun every test. The core provisioning policy must remain unchanged.
The API authorization registry must grant `writer:provision` only to the
authenticated Owner role. Agents and ordinary writer tokens must not receive
this scope.
## Privilege boundary
The API service must remain non-root.
Install a root-owned sudoers rule that allows the API service identity to run
only these exact commands with `sudo -n --`:
```text
/usr/local/libexec/aios-grant-signer-adapter
/usr/local/libexec/aios-grant-validator
/usr/local/libexec/aios-task-writer-provisioner
```
Do not allow:
```text
/bin/sh
/bin/bash
/usr/bin/env
/usr/bin/git
wildcard executable paths
SETENV
arbitrary environment preservation
```
The governed tools themselves must validate their `--grant-dir` and
`--result-json` paths. Their binaries must be root-owned, non-symlink, regular,
executable and not group/world writable.
The signer must access its private signing material only through a runtime
secret reference. No signing key or token may enter Git, API output, the
provisioning receipt or systemd environment text.
## Runtime paths and permissions
Create:
```text
/var/lib/aios-gitops/authorized-writer-bridge
/var/lib/aios-gitops/authorized-writer-bridge/receipts
/var/lib/aios-gitops/authorized-writer-bridge/locks
```
Required properties:
```yaml
owner: API service identity
mode: "0700"
symlink: false
persistent: true
```
The directory contains redacted receipts only. Temporary signed grant
directories are deleted after success or failure.
## Required staging environment
Configure the API service:
```text
AIOS_GITOPS_REPOSITORY_IDENTITY=aios/aios
AIOS_GITOPS_AUTHORITY_GIT_DIR=/data/aios-gitops/repo.git
AIOS_GITOPS_AUTHORIZED_WRITER_STATE_ROOT=/var/lib/aios-gitops/authorized-writer-bridge
AIOS_GITOPS_GRANT_SIGNER_ADAPTER=/usr/local/libexec/aios-grant-signer-adapter
AIOS_GITOPS_GRANT_VALIDATOR=/usr/local/libexec/aios-grant-validator
AIOS_GITOPS_TASK_WRITER_PROVISIONER=/usr/local/libexec/aios-task-writer-provisioner
AIOS_GITOPS_OWNER_PROVISION_SCOPE=writer:provision
AIOS_GITOPS_GOVERNED_USE_SUDO=true
AIOS_GITOPS_SUDO_PATH=/usr/bin/sudo
```
Use `aios_authorized_task_writer_api.py` as the API entrypoint. It delegates
all pre-existing routes and adds only the new endpoint.
## Pre-merge tests
Run from the repository root:
```bash
python3 -m unittest -v tests/test_authorized_task_writer_provisioning.py
python3 scripts/smoke_authorized_task_writer_provisioning.py
python3 -m py_compile \
active-runtime/app/authorized_task_writer_provisioning.py \
active-runtime/app/aios_authorized_task_writer_api.py
```
Required result:
```yaml
unit_tests:
total: 9
failures: 0
errors: 0
smoke: PASS
py_compile: PASS
```
Also run all existing GitOps API and isolated writer tests affected by:
```text
aios_chatgpt_actions_api
aios_isolated_writer_api
isolated_writer_authz
isolated_writer_token_registry
single_owner_governance
OpenAPI schema validation
```
Any regression blocks merge.
## Staging API acceptance
### Negative authentication
An Agent or non-Owner credential calling the endpoint must receive:
```yaml
http_status: 403
error: OWNER_AUTHORIZATION_REQUIRED
```
### Negative scope tests
Requests enabling any of the following must be denied before signer execution:
```text
production_change_authorized=true
tag_change_authorized=true
baseline_change_authorized=true
main_write_authorized=true
allowed_operations containing tag/main/baseline/admin operations
missing required denied operation
one_time_use=false
expiry over 1800 seconds
```
### Drift and conflict tests
Verify:
```yaml
wrong_base_oid:
http_status: 409
error: AUTHORIZED_BASE_OID_DRIFT
existing_unreceipted_target_branch:
http_status: 409
error: EXISTING_BRANCH_OID_CONFLICT
```
### Governed tool failures
Independently inject and verify fail-closed behavior for:
```text
signer unavailable
signer signature absent
validator signature_valid=false
validator signer_trusted=false
validator plan expired
provisioner grant_consumed=false
provisioner writer scope enables main/tag/baseline/force push
any required negative policy test false
```
None may create or modify a Git ref.
## Live V2R1C6 provisioning request
After protected merge and API deployment, authenticate as the Owner and submit
exactly:
```json
{
"task_id": "V2R1C6_TASK_SCOPED_WRITER_PROVISIONING",
"repository": "aios/aios",
"base_ref": "refs/heads/main",
"base_oid": "875d41c3101093149414292067fd700dc214d4d3",
"target_branch": "agent/v2r1c6-mainline-reconciliation-20260805",
"owner_authorization_reference": "chat-owner-authorization:v2r1c6:20260806",
"one_time_use": true,
"expires_in_seconds": 900,
"production_change_authorized": false,
"tag_change_authorized": false,
"baseline_change_authorized": false,
"main_write_authorized": false,
"allowed_operations": [
"create_worktree",
"create_target_branch_if_absent",
"commit_to_target_branch",
"push_target_branch"
],
"denied_operations": [
"direct_main_push",
"force_push",
"tag_create",
"tag_update",
"tag_delete",
"branch_delete",
"repository_admin",
"baseline_write",
"production_deploy",
"production_marker_write",
"history_rewrite"
]
}
```
This request must be executed by the localhost-only bootstrap instance before
the bridge code advances `main`. The endpoint must never infer “current main”.
If the Agent chooses the alternative merge-first path, it must stop after the
merge and obtain an exact Owner amendment for the new merge OID. It must not
silently replace the approved OID.
## Success response
Require:
```yaml
status: READY_FOR_CODE_AUTHORING
grant_consumed: true
branch: agent/v2r1c6-mainline-reconciliation-20260805
branch_oid: <exact amended authorized base OID>
base_oid: <same exact OID>
main_unchanged: true
writer_scope:
workspace_role: task-scoped-isolated-writer
writes_enabled: true
allowed_branch: agent/v2r1c6-mainline-reconciliation-20260805
allowed_base_oid: <exact amended authorized base OID>
allowed_routes:
- POST /api/commit-file
allow_main_write: false
allow_tag_write: false
allow_baseline_write: false
allow_force_push: false
```
All seven negative policy results must be true.
Submit the identical request a second time. It must return the same grant and
branch attestation with:
```yaml
idempotent: true
```
It must not run the provisioner again.
## Production postchecks
After API deployment and writer provisioning, independently verify:
```yaml
existing_routes_regression: PASS
new_endpoint: PASS
main_ref: expected reviewed merge commit
V2R1C5_tag_oid: 9e65bd59baea392294c0f253046d3ab116f63145
V2R1C5_peeled_target: aa60e891547abd7e215af7cf2bf3f21fd23541da
V2R1C5_baseline_unchanged: true
new_release_tag_created: false
production_baseline_changed: false
```
The deployment of this control-plane bridge does not authorize a V2R1C6
production release. It only enables scoped code authoring.
## Rollback
If the new API route fails:
1. Stop accepting new provisioning calls.
2. Restore the previous API entrypoint.
3. Restart and verify all existing routes.
4. Preserve receipts and logs.
5. Revoke any task writer capability already minted.
6. Do not delete a branch or rewrite a tag.
7. Revert the protected-main bridge commit through an ordinary reviewed revert
PR if source rollback is required.
## Required final Agent report
```yaml
STATUS: COMPLETED | PARTIALLY_COMPLETED | BLOCKED | FAILED
INTEGRATION:
branch:
commit:
protected_pr:
merge_commit:
files_changed:
TESTS:
unit:
smoke:
py_compile:
regression:
fault_injection:
DEPLOYMENT:
service:
entrypoint:
runtime_status:
endpoint_status:
rollback_verified:
PROVISIONING:
owner_authorization_reference:
approval_plan_id:
grant_id:
grant_digest_sha256:
signing_key_identity:
branch:
branch_oid:
workspace_path:
grant_consumed:
idempotent_replay:
CONSISTENCY:
main_oid:
V2R1C5_tag_oid:
V2R1C5_peeled_target_oid:
baseline_version:
baseline_changed:
tags_changed:
SECRETS:
runtime_secret_reference_used:
plaintext_secret_logged:
plaintext_secret_committed:
BLOCKERS:
GOVERNANCE:
force_push_used: false
tag_rewritten: false
direct_main_push: false
secrets_committed: false
owner_approval_respected: true
```
@@ -0,0 +1,401 @@
# Agent Deployment Requirements: Owner Authorization Broker V2
## Objective
Deploy the audited Owner Authorization Broker that closes the bootstrap gap
without issuing a new general-purpose Owner credential and without modifying
`/data/aios-gitops/runtime/secrets/gitops.tokens`.
The broker reuses one existing, non-wildcard Owner credential only to
authenticate the Owner identity through one pre-existing exact scope. It then
records an immutable task authorization event and directly invokes the
existing signed grant -> validator -> one-shot provisioner chain.
Production deployment of V2R1C6 is **not** authorized by this task.
## Fixed bindings
```yaml
task_id: V2R1C6_TASK_SCOPED_WRITER_PROVISIONING
repository: aios/aios
base_ref: refs/heads/main
base_oid: 875d41c3101093149414292067fd700dc214d4d3
target_branch: agent/v2r1c6-mainline-reconciliation-20260805
production_change_authorized: false
tag_change_authorized: false
baseline_change_authorized: false
main_write_authorized: false
```
The existing V2R1C5 objects are frozen:
```yaml
tag_oid: 9e65bd59baea392294c0f253046d3ab116f63145
peeled_target_oid: aa60e891547abd7e215af7cf2bf3f21fd23541da
baseline_version: V2R1C5
baseline_status: PRODUCTION_DEPLOYED
```
## 1. Package verification
Verify `MANIFEST.sha256` before installation. Reject any mismatch.
Run:
```bash
python3 -m unittest -v \
tests/test_authorized_task_writer_provisioning.py \
tests/test_owner_authorization_broker.py
python3 scripts/smoke_authorized_task_writer_provisioning.py
python3 scripts/smoke_owner_authorization_broker.py
python3 -m py_compile \
active-runtime/app/authorized_task_writer_provisioning.py \
active-runtime/app/aios_authorized_task_writer_api.py \
active-runtime/app/owner_authorization_broker.py \
active-runtime/app/aios_owner_authorization_broker_api.py
```
All checks must pass.
## 2. Existing Owner credential selection
Read only the non-secret registry metadata. Select one existing credential
that satisfies all of the following:
```yaml
role: owner
wildcard_scope: false
account: exact known Owner identity
contains_exact_intake_scope: true
```
Use one existing exact scope such as `commit:create` or `file:write`. Prefer
the narrowest scope already assigned to the intended Owner.
Do not:
- use the historical wildcard Owner credential;
- add `writer:provision` to any bearer token;
- create or modify a bearer credential;
- print, copy or commit bearer-token material.
Record only:
```yaml
owner_account: <non-secret account>
credential_id: <non-secret ID, when available>
intake_scope: <one exact existing scope>
scope_readback_available: true|false
```
If `handler.need()` returns the exact scope set, configure the Owner account
allowlist and leave the credential ID allowlist empty.
If scope metadata is not returned, configure the exact non-secret credential
ID allowlist. If neither exact scope metadata nor a stable non-secret
credential ID can be proven, stop:
```yaml
STATUS: BLOCKED
REASON: OWNER_CREDENTIAL_SCOPE_NOT_PROVABLE
```
## 3. Staging installation
Install the four application modules in a root-owned staging directory:
```text
authorized_task_writer_provisioning.py
aios_authorized_task_writer_api.py
owner_authorization_broker.py
aios_owner_authorization_broker_api.py
```
The directory and files must not be group/world writable.
Prepare environment values:
```yaml
AIOS_GITOPS_OWNER_AUTHORIZATION_BROKER_ENABLED: "true"
AIOS_GITOPS_OWNER_AUTHORIZATION_INTAKE_SCOPES: <one exact existing scope>
AIOS_GITOPS_OWNER_AUTHORIZATION_ALLOWED_IDENTITIES: <exact owner account>
AIOS_GITOPS_OWNER_AUTHORIZATION_ALLOWED_CREDENTIAL_IDS: <exact ID or empty>
AIOS_GITOPS_OWNER_AUTHORIZATION_MAX_EXPIRY_SECONDS: "900"
AIOS_GITOPS_OWNER_AUTHORIZATION_STATE_ROOT:
/var/lib/aios-gitops/owner-authorization-broker
```
Use the existing signer, validator and provisioner paths. Do not change their
root governance configuration or disable signature checks.
Start a localhost-only staging instance on a temporary port. Do not replace
the production service yet.
## 4. Authentication fail-closed tests
The staging endpoint is:
```text
POST /api/governance/owner-authorizations
```
Required tests:
```yaml
unauthenticated:
expected: 401
wrong_role:
expected: 403
wrong_scope:
expected: 403
wildcard_owner:
expected: 403
error: WILDCARD_OWNER_CREDENTIAL_REJECTED
unallowlisted_owner:
expected: 403
error: OWNER_IDENTITY_NOT_ALLOWED
production_capability_true:
expected: 403
error: PRODUCTION_CAPABILITY_REJECTED
```
A connection failure, 404 or 500 is not an authorization-policy pass.
## 5. Formal provisioning request
Use the selected existing non-wildcard Owner credential through its runtime
secret reference. Never place the bearer token in the request body, command
line, shell history, output report or Git.
Request:
```json
{
"action": "AUTHORIZE_TASK_WRITER_PROVISIONING",
"task_id": "V2R1C6_TASK_SCOPED_WRITER_PROVISIONING",
"repository": "aios/aios",
"base_ref": "refs/heads/main",
"base_oid": "875d41c3101093149414292067fd700dc214d4d3",
"target_branch": "agent/v2r1c6-mainline-reconciliation-20260805",
"one_time_use": true,
"expires_in_seconds": 900,
"production_change_authorized": false,
"tag_change_authorized": false,
"baseline_change_authorized": false,
"main_write_authorized": false,
"allowed_operations": [
"create_worktree",
"create_target_branch_if_absent",
"commit_to_target_branch",
"push_target_branch"
],
"denied_operations": [
"direct_main_push",
"force_push",
"tag_create",
"tag_update",
"tag_delete",
"branch_delete",
"repository_admin",
"baseline_write",
"production_deploy",
"production_marker_write",
"history_rewrite"
]
}
```
The Owner authorization reference is generated by the broker. The caller must
not submit or invent one.
Expected result:
```yaml
status: READY_FOR_CODE_AUTHORING
owner_authorization_event_id: generated
owner_authorization_reference: generated
owner_authorization_digest_sha256: generated
owner_authorization_consumed: true
approval_plan_id: generated
grant_digest_sha256: generated
grant_consumed: true
branch_oid: 875d41c3101093149414292067fd700dc214d4d3
writer_scope:
workspace_role: task-scoped-isolated-writer
writes_enabled: true
allowed_branch: agent/v2r1c6-mainline-reconciliation-20260805
allowed_base_oid: 875d41c3101093149414292067fd700dc214d4d3
allow_main_write: false
allow_tag_write: false
allow_baseline_write: false
allow_force_push: false
```
## 6. Independent post-provisioning verification
Verify the Authority ref:
```bash
git --git-dir=/data/aios-gitops/repo.git rev-parse \
refs/heads/agent/v2r1c6-mainline-reconciliation-20260805
```
It must equal:
```text
875d41c3101093149414292067fd700dc214d4d3
```
Verify `main` remains unchanged, V2R1C5 tag object and peeled target remain
unchanged, and the Production Baseline remains V2R1C5.
Verify the broker audit event under:
```text
/var/lib/aios-gitops/owner-authorization-broker/events/
```
Requirements:
```yaml
status: CONSUMED
wildcard_scope_used: false
grant_consumed: true
permissions:
directory: "0700"
json_files: "0600"
```
Do not return bearer-token material in the report.
## 7. Commit the broker through the new writer
After `READY_FOR_CODE_AUTHORING`, use only the writer's
`POST /api/commit-file` route to put this code bundle on:
```text
agent/v2r1c6-mainline-reconciliation-20260805
```
Commit the bridge and broker as separate intentional commits when practical:
1. Owner-authorized task-writer bridge.
2. Audited Owner authorization broker and deployment docs.
Do not merge to `main` directly. Create a protected PR after branch commits
and tests are complete.
## 8. Permanent service deployment
Only after protected review and merge:
1. Install the reviewed files from the merge commit.
2. Apply a reviewed systemd drop-in based on the provided example.
3. Start the wrapper entrypoint:
`aios_owner_authorization_broker_api.py`.
4. Verify all existing API routes still work.
5. Verify the broker remains localhost/reverse-proxy constrained as intended.
6. Re-run authentication and policy tests.
7. Remove the temporary staging instance.
This permanent bridge deployment does not authorize a V2R1C6 production tag,
Baseline update or application deployment.
## 9. Fail-closed conditions
Stop with `BLOCKED` for any of the following:
- Authority `main` no longer equals the fixed base OID.
- Target branch exists at a different OID.
- Only a wildcard Owner credential can authenticate.
- Exact credential scope cannot be proven.
- Owner identity is not explicitly allow-listed.
- Signer, validator or provisioner self-test fails.
- Grant is not marked one-time and consumed.
- Writer allows main, tag, Baseline, force-push or branch-delete operations.
- Audit event contains sensitive material.
- V2R1C5 tag or Baseline changes.
## 10. Required Agent result
```yaml
STATUS: READY_FOR_CODE_AUTHORING | BLOCKED | FAILED
PACKAGE:
manifest_verified:
unit_tests:
smoke_tests:
py_compile:
OWNER_CREDENTIAL:
account:
credential_id:
intake_scope:
exact_scope_proven:
wildcard_used: false
token_logged: false
BROKER:
staging_listener:
endpoint_status:
event_id:
authorization_reference:
authorization_digest_sha256:
event_consumed:
audit_permissions_valid:
GRANT:
approval_plan_id:
grant_digest_sha256:
signer_identity:
signature_valid:
one_time_use:
consumed:
BRANCH:
name:
oid:
base_oid:
ref_exists:
worktree_path:
worktree_clean:
WRITER:
status:
writes_enabled:
allowed_branch:
allowed_base_oid:
allowed_routes:
main_write_denied:
tag_write_denied:
baseline_write_denied:
force_push_denied:
POSTCHECK:
main_unchanged:
V2R1C5_tag_unchanged:
baseline_unchanged:
REPOSITORY_INTEGRATION:
commit_oids:
protected_pr:
merged:
permanent_bridge_deployed:
BLOCKERS:
GOVERNANCE:
force_push_used: false
tag_rewritten: false
direct_main_push: false
secrets_committed: false
owner_approval_respected: true
```
+562
View File
@@ -0,0 +1,562 @@
# Agent execution requirements: V2R1C6 task identity overlay
## Required result
The Agent must return one of:
- `READY_FOR_CODE_AUTHORING`
- `BLOCKED`
- `FAILED`
The Agent may not report success until the task branch, isolated worktree and
task writer are independently verified.
This stage is **not** a production deployment. It must not modify `main`,
V2R1C5, tags, Registry, Production Baseline or production services.
## Fixed authority
```yaml
task_id: V2R1C6_TASK_SCOPED_WRITER_PROVISIONING
repository: aios/aios
base_ref: refs/heads/main
base_oid: 875d41c3101093149414292067fd700dc214d4d3
target_branch: agent/v2r1c6-mainline-reconciliation-20260805
owner_identity: aiman
one_time_use: true
maximum_ttl_seconds: 900
production_change_authorized: false
tag_change_authorized: false
baseline_change_authorized: false
main_write_authorized: false
```
## 1. Verify package
Verify the externally supplied package against the published SHA-256 values and
then verify every entry in `MANIFEST.sha256`.
Do not synthesize a missing archive. Use either the verified ZIP or TAR.GZ.
Required checks:
```yaml
package_hash_match: true
manifest_entries_match: true
unit_tests_pass: true
smoke_tests_pass: true
py_compile_pass: true
```
## 2. Create root-owned staging
Extract to a fresh directory such as:
```text
/var/lib/aios-gitops/staging/v2r1c6-task-identity-overlay-<nonce>
```
Requirements:
```yaml
owner: root
group: root
directory_mode: "0700"
files_not_world_writable: true
symlinks_rejected: true
```
Do not overwrite the active runtime tree.
## 3. Re-run target-host tests
From the staging root:
```bash
python3 -m unittest discover -s tests -v
python3 scripts/smoke_authorized_task_writer_provisioning.py
python3 scripts/smoke_owner_authorization_broker.py
python3 scripts/smoke_task_identity_overlay.py
python3 -m py_compile \
active-runtime/app/authorized_task_writer_provisioning.py \
active-runtime/app/owner_authorization_broker.py \
active-runtime/app/task_identity_overlay.py \
active-runtime/app/aios_authorized_task_writer_api.py \
active-runtime/app/aios_owner_authorization_broker_api.py \
active-runtime/app/aios_task_identity_overlay_api.py \
active-runtime/app/aios_task_identity_overlay_cli.py \
scripts/call_task_identity_overlay.py
```
Any failure is `BLOCKED`; do not proceed.
## 4. Prestart Git and production freeze checks
Read from the authority repository:
```bash
git --git-dir=/data/aios-gitops/repo.git rev-parse refs/heads/main
git --git-dir=/data/aios-gitops/repo.git \
show-ref --verify --quiet \
refs/heads/agent/v2r1c6-mainline-reconciliation-20260805
git --git-dir=/data/aios-gitops/repo.git rev-parse refs/tags/V2R1C5
git --git-dir=/data/aios-gitops/repo.git rev-parse 'refs/tags/V2R1C5^{}'
```
Expected:
```yaml
main_oid: 875d41c3101093149414292067fd700dc214d4d3
target_branch_exists: false
V2R1C5_tag_oid: 9e65bd59baea392294c0f253046d3ab116f63145
V2R1C5_peeled_target: aa60e891547abd7e215af7cf2bf3f21fd23541da
```
Read Baseline and require:
```yaml
version: V2R1C5
status: PRODUCTION_DEPLOYED
target_oid: aa60e891547abd7e215af7cf2bf3f21fd23541da
evidence_closure_oid: 5e2cf8276f5093f3352b9a3252c701a44c5ab807
```
If `main` has drifted, return:
```yaml
STATUS: MANUAL_OWNER_REVIEW_REQUIRED
REASON: AUTHORIZED_BASE_OID_DRIFT
```
If the target branch already exists at another OID, return:
```yaml
STATUS: MANUAL_OWNER_REVIEW_REQUIRED
REASON: EXISTING_BRANCH_OID_CONFLICT
```
Do not delete, reset or force-push the branch.
## 5. Discover active secret references
The overlay does not read or modify `gitops.tokens`, but its issuer must compare
the new random secret against all active bearer secret references.
Discover references without printing secret contents. At minimum include the
currently active ChatGPT/GitOps bearer secret reference and any other bearer
files consumed by the active API.
Pass them as a comma-separated list in:
```text
AIOS_GITOPS_IDENTITY_OVERLAY_COLLISION_SECRET_REFS
```
Evidence must contain only paths and SHA-256 match counts, not token values.
If active secret references cannot be enumerated, return:
```yaml
STATUS: BLOCKED
REASON: ACTIVE_SECRET_REFERENCE_SET_UNRESOLVED
```
## 6. Install the transient systemd service
Use the provided template:
```text
deploy/systemd/aios-v2r1c6-task-identity-overlay.service.example
```
Substitute only:
- verified root-owned staging directory;
- verified active secret-ref path list.
Do not change task ID, repository, base OID, target branch or denied
capabilities.
The service must listen only on:
```text
127.0.0.1:19864
```
Before start, run:
```bash
systemd-analyze verify \
/etc/systemd/system/aios-v2r1c6-task-identity-overlay.service
```
Then:
```bash
sudo systemctl daemon-reload
sudo systemctl start aios-v2r1c6-task-identity-overlay.service
```
Confirm:
```bash
curl --fail --silent http://127.0.0.1:19864/healthz
```
Expected:
```yaml
service: aios-task-identity-overlay
status: active
global_token_registry_used: false
loopback_only: true
```
Do not expose the port externally.
## 7. Issue the one-time overlay
Use the root-only CLI. The authorization assertion is an audit label for this
already approved Owner task; it must not be described as an external signed
event.
Example:
```bash
sudo env \
PYTHONPATH=<staging>/active-runtime/app \
AIOS_GITOPS_IDENTITY_OVERLAY_STATE_ROOT=/var/lib/aios-gitops/task-identity-overlays \
AIOS_GITOPS_IDENTITY_OVERLAY_SECRET_ROOT=/run/aios-gitops/task-identity-overlays \
AIOS_GITOPS_IDENTITY_OVERLAY_TASK_ID=V2R1C6_TASK_SCOPED_WRITER_PROVISIONING \
AIOS_GITOPS_IDENTITY_OVERLAY_OWNER_IDENTITY=aiman \
AIOS_GITOPS_IDENTITY_OVERLAY_BASE_REF=refs/heads/main \
AIOS_GITOPS_IDENTITY_OVERLAY_BASE_OID=875d41c3101093149414292067fd700dc214d4d3 \
AIOS_GITOPS_IDENTITY_OVERLAY_TARGET_BRANCH=agent/v2r1c6-mainline-reconciliation-20260805 \
AIOS_GITOPS_IDENTITY_OVERLAY_COLLISION_SECRET_REFS=<verified-secret-ref-list> \
python3 <staging>/active-runtime/app/aios_task_identity_overlay_cli.py \
issue \
--ttl-seconds 900 \
--authorization-assertion-ref owner-session:V2R1C6:approved-20260806 \
--operator-identity <audited-agent-identity> \
--result-json /run/aios-gitops/v2r1c6-overlay-issued.json
```
The result must contain:
```yaml
status: OVERLAY_ACTIVE
idempotent: false
overlay_id: present
secret_ref: under /run/aios-gitops/task-identity-overlays
expires_at: within 900 seconds
```
It must not contain a token.
Verify:
```yaml
secret_file_mode: "0600"
metadata_contains_raw_token: false
token_hash_unique_across_configured_secret_refs: true
```
## 8. Negative tests
Before the valid request:
1. No Authorization header must return HTTP 401.
2. A random wrong bearer must return HTTP 401.
3. A request with `production_change_authorized=true` must be rejected.
4. A request with another base OID or branch must be rejected.
5. The valid overlay must remain `ACTIVE` after pre-auth request-validation
failures.
Do not use the global Owner or ChatGPT bearer credential for this endpoint.
## 9. Send the valid request
Use:
```text
deploy/v2r1c6-task-writer-authorization-request.json
```
Use the provided client so the bearer is not placed in argv, shell history or
logs:
```bash
sudo python3 <staging>/scripts/call_task_identity_overlay.py \
--secret-ref <secret_ref_from_issue_result> \
--request-json <staging>/deploy/v2r1c6-task-writer-authorization-request.json \
--url http://127.0.0.1:19864/api/governance/owner-authorizations \
--result-json /var/lib/aios-gitops/v2r1c6-overlay-provision-result.json
```
Required result:
```yaml
http_status: 200
status: READY_FOR_CODE_AUTHORING
owner_authorization_consumed: true
grant_consumed: true
identity_overlay_consumed: true
identity_overlay_secret_removed: true
branch: agent/v2r1c6-mainline-reconciliation-20260805
branch_oid: 875d41c3101093149414292067fd700dc214d4d3
```
The result must not include:
- bearer token;
- private key;
- signer secret;
- raw signature;
- credential value.
## 10. Independent acceptance
Verify the branch:
```bash
git --git-dir=/data/aios-gitops/repo.git \
rev-parse refs/heads/agent/v2r1c6-mainline-reconciliation-20260805
```
It must equal:
```text
875d41c3101093149414292067fd700dc214d4d3
```
Verify worktree evidence returned by the provisioner:
```yaml
branch: agent/v2r1c6-mainline-reconciliation-20260805
head_oid: 875d41c3101093149414292067fd700dc214d4d3
detached_head: false
worktree_clean: true
writer_status: active
writes_enabled: true
allowed_routes:
- POST /api/commit-file
allow_main_write: false
allow_tag_write: false
allow_baseline_write: false
allow_force_push: false
```
Run policy negative tests against the task writer and require explicit policy
denials for:
- main write;
- another branch;
- tag create/update;
- Baseline write;
- force push;
- branch delete.
Network errors, 404, 500 and stopped-service failures do not count as policy
denials.
## 11. Stop and clean up the overlay
After `READY_FOR_CODE_AUTHORING`:
```bash
sudo systemctl stop aios-v2r1c6-task-identity-overlay.service
sudo systemctl disable aios-v2r1c6-task-identity-overlay.service || true
sudo rm -f /etc/systemd/system/aios-v2r1c6-task-identity-overlay.service
sudo systemctl daemon-reload
```
Verify:
```yaml
overlay_listener_stopped: true
overlay_secret_removed: true
global_token_registry_unchanged: true
disabled_owner_record_unchanged: true
active_chatgpt_bot_record_unchanged: true
audit_metadata_retained: true
```
The root-owned staging directory may be retained until the code is committed,
then removed after checksum-confirmed repository integration.
## 12. Commit the developed code through the task writer
Only after writer acceptance, use `POST /api/commit-file` to commit the
verified package contents to:
```text
agent/v2r1c6-mainline-reconciliation-20260805
```
Commit in reviewable groups:
1. authorized task-writer bridge;
2. Owner authorization broker;
3. task identity overlay;
4. tests, docs, OpenAPI and deployment templates.
Do not include:
- package archives;
- runtime tokens;
- `/run` or `/var/lib` state;
- test result files containing host paths;
- secret references containing values;
- generated credentials.
Run the complete 40-test suite from the branch after commits.
## 13. Protected PR
Create a protected PR from:
```text
agent/v2r1c6-mainline-reconciliation-20260805
```
to:
```text
main
```
The PR must state:
- this change closes Owner authorization bootstrap gaps;
- the overlay is localhost-only and one-time;
- the global token registry is not modified;
- no production release is authorized;
- V2R1C5 tag and Baseline remain unchanged.
The Agent must not merge without the repository's normal protected review.
## 14. Required Agent report
```yaml
STATUS:
READY_FOR_CODE_AUTHORING | BLOCKED | FAILED
PACKAGE:
zip_sha256:
tar_gz_sha256:
manifest_entries:
manifest_matched:
TESTS:
unit_total:
unit_passed:
smoke_bridge:
smoke_broker:
smoke_overlay:
py_compile:
PRESTART:
main_oid:
target_branch_previously_absent:
V2R1C5_tag_oid:
V2R1C5_peeled_target:
baseline_unchanged:
OVERLAY:
overlay_id:
status:
secret_ref:
secret_file_mode:
token_hash_unique:
global_registry_used: false
listener:
expires_at:
consumed:
secret_removed:
NEGATIVE_TESTS:
unauthenticated_401:
wrong_token_401:
production_capability_denied:
wrong_base_denied:
wrong_branch_denied:
OWNER_AUTHORIZATION:
event_id:
reference:
consumed:
GRANT:
approval_plan_id:
grant_id:
grant_digest_sha256:
signature_valid:
consumed:
BRANCH:
name:
oid:
base_oid:
ref_exists:
worktree_path:
worktree_clean:
WRITER:
status:
writes_enabled:
allowed_branch:
allowed_base_oid:
allowed_routes:
main_write_denied:
tag_write_denied:
baseline_write_denied:
force_push_denied:
REPOSITORY_INTEGRATION:
commits:
tests_after_commit:
protected_pr:
merged: false
POSTCHECK:
main_unchanged:
V2R1C5_tag_unchanged:
baseline_unchanged:
global_token_registry_unchanged:
overlay_listener_stopped:
overlay_secret_removed:
SECRETS:
plaintext_logged: false
plaintext_committed: false
BLOCKERS:
GOVERNANCE:
force_push_used: false
tag_rewritten: false
direct_main_push: false
secrets_committed: false
owner_approval_respected: true
```
## Fail-closed conditions
Return `BLOCKED` and stop on any of:
- authoritative main OID drift;
- target branch conflict;
- inability to enumerate active secret references;
- secret hash collision;
- overlay not bound to the exact task/base/branch;
- non-loopback listener;
- raw secret in result or logs;
- Owner Authorization Broker event not created;
- signer/validator/provisioner failure;
- grant not consumed;
- branch OID mismatch;
- writer exposes main/tag/Baseline/force-push permissions;
- V2R1C5 tag or Baseline changes.
Never fall back to the shared `gitops-write-bot.token`, a wildcard credential,
manual `gitops.tokens` edits or a direct Git ref write.
+9
View File
@@ -0,0 +1,9 @@
{
"schema": "aios.v2r1c7.engineering-build-receipt.v1",
"version": "V2R1C7",
"base_oid": "b356dec2ac5bd2a5f721aa73f8fd1902f9c65632",
"production_deployed": false,
"tag_created": false,
"baseline_changed": false,
"deployable": true
}
+21
View File
@@ -0,0 +1,21 @@
bec7da1adb75bf5feaa63c60ec6d03adf15d7508742a486207033d000f72d4ff AGENT_DEPLOYMENT_REQUIREMENTS.md
251b46e51ce09aa2fd843bee48ef24a5608f31d24c599841f086ceb25c1a1715 README.md
783f68ec4f00a11d43ceed536196bfedccaf6614f6db9805f7b36d24e5fa7bc2 TEST_RESULTS.json
9208325e5b2e6f5ba6b64e000bc21d14d9f2f214e3ad33dd79857e85218fd6c2 VERSION
c4b24b65195077716983acdc8611f705324413a74453858994ba51541957afd4 active-runtime/app/aios_authorized_task_writer_api.py
8a1935fa098d47ac02ce51bbd9fd5fad7cf9ab9374ea3de5f630679b96681d84 active-runtime/app/authorized_task_writer_provisioning.py
0997ab8332985c7124f334c4e49735d8833326578acc61a61a389bf792424de3 active-runtime/app/deployment_state.py
dd3f8b1cab88e005a291f077807873c381e8ca92fe863dc05817925f0dff9613 active-runtime/app/v2r1c6_privacy_tools.py
9e45025c41068cafdea2f360edfc4146a5459b0a6bcfbf1403206fc0161e7bac active-runtime/app/v2r1c7_health_service.py
c96ba560d69f6c2eb700b9f2b42a32e394dadd00dffb23bbefdb9fb3e1571c0d deploy/aios-v2/systemd/aios-v2r1c7-production.service
a05d4e8038fdb866ce17d4fb8a54798a268a1e8d44ff2cbc559069fa6ab2a667 deploy/systemd/aios-gitops-model-api-authorized-writer-bridge.conf.example
a60dac51802bff67c2cfff61acdeb0a3a05d74a4f6cb47e3f78295ceb5c6acd9 docs/gitops/authorized-task-writer-provisioning.md
2690f76452bceddda90c728278c600bcebd533db24776e2966423c5a8655e4b5 docs/gitops/openapi-authorized-task-writer-fragment.yaml
560e5d9af8a8e2ec60ec31cf8659a7982b93ea09b7adb486b738ea2fba460d70 release-manifests/V2R1C7.json
81a423c7814f81cb888b60a538c153e549b2be9099c881c968d55ae84efddc57 scripts/build_v2r1c7_release_artifact.py
64c78dabffb05c8851f87b2a27c43bd1f21728dca586397d03a6bc22906a906c scripts/probe_v2r1c7_health.py
5419dc7de0f50908ad0ef1dac74172fce811bb7b45f314a6e1c1bb9ce618324d scripts/rollback_v2r1c7_release.sh
54e2845d2c3c026323ad45515e082d661e3b461feb3bfefedf5adc52db2d7b78 scripts/smoke_authorized_task_writer_provisioning.py
32a860e28d5afc105f413d878a13fb8600fa74874dbd3bc4c031ef3efa17896a scripts/verify_v2r1c7_release.py
3cce2b744326fe12b89bcad3cd352c77738950964bf8519e72cb5e9c7a5dfbfc tests/test_authorized_task_writer_provisioning.py
8f3153c625241f60b65109eeac045e8aed7661d95f3a10f0b32ba3e0e9ac0a85 versions/V2R1C7.json
+25
View File
@@ -0,0 +1,25 @@
{
"git_branch_created": false,
"git_commit_created": false,
"production_mutation": false,
"py_compile": {
"files": [
"active-runtime/app/authorized_task_writer_provisioning.py",
"active-runtime/app/aios_authorized_task_writer_api.py"
],
"status": "PASS"
},
"schema": "aios.authorized-task-writer-bridge-test-report.v1",
"secrets_included": false,
"smoke": {
"command": "python3 scripts/smoke_authorized_task_writer_provisioning.py",
"status": "PASS"
},
"status": "PASS",
"unit_tests": {
"command": "python3 -m unittest -v tests/test_authorized_task_writer_provisioning.py",
"errors": 0,
"failures": 0,
"total": 9
}
}
+1 -1
View File
@@ -1 +1 @@
V2R1C2
V2R1C7
+201
View File
@@ -0,0 +1,201 @@
"""Fail-closed Agent Pull Runner primitives for AIOS autonomy v1.
The runner owns task discovery and reporting only. It never accepts shell
strings: executors receive argv, an allowlisted cwd and an allowlisted env.
Durable state is intentionally small so restart recovery and duplicate
suppression can be verified without coupling the runner to one database.
"""
from __future__ import annotations
import json
import os
import random
import shutil
import subprocess
import tempfile
import time
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Callable, Mapping, Protocol, Sequence
class RunnerError(RuntimeError):
"""Raised when a runner contract would be violated."""
@dataclass(frozen=True)
class RunnerConfig:
worker_id: str
capabilities: tuple[str, ...] = ()
long_poll_seconds: int = 30
fallback_seconds: int = 60
jitter_seconds: int = 5
max_backoff_seconds: int = 300
executable_allowlist: tuple[str, ...] = ()
working_directory_allowlist: tuple[str, ...] = ()
environment_allowlist: tuple[str, ...] = ()
@dataclass
class RunnerState:
inflight: dict[str, dict[str, Any]] = field(default_factory=dict)
terminal: dict[str, dict[str, Any]] = field(default_factory=dict)
class RunnerTransport(Protocol):
def register_worker(self, worker_id: str, capabilities: Sequence[str]) -> Mapping[str, Any]: ...
def heartbeat(self, worker_id: str) -> Mapping[str, Any]: ...
def long_poll(self, worker_id: str, timeout_seconds: int) -> Mapping[str, Any] | None: ...
def poll(self, worker_id: str) -> Mapping[str, Any] | None: ...
def claim(self, worker_id: str, task: Mapping[str, Any]) -> Mapping[str, Any]: ...
def renew(self, worker_id: str, task_id: str) -> Mapping[str, Any]: ...
def report_progress(self, worker_id: str, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]: ...
def report_result(self, worker_id: str, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]: ...
def report_evidence(self, worker_id: str, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]: ...
class JsonStateStore:
"""Atomic JSON state store for restart recovery."""
def __init__(self, path: str | os.PathLike[str]) -> None:
self.path = Path(path)
def load(self) -> RunnerState:
if not self.path.exists():
return RunnerState()
value = json.loads(self.path.read_text(encoding="utf-8"))
if not isinstance(value, dict):
raise RunnerError("runner state must be an object")
return RunnerState(
inflight=dict(value.get("inflight", {})),
terminal=dict(value.get("terminal", {})),
)
def save(self, state: RunnerState) -> None:
self.path.parent.mkdir(parents=True, exist_ok=True)
fd, name = tempfile.mkstemp(prefix=self.path.name + ".", dir=self.path.parent)
try:
with os.fdopen(fd, "w", encoding="utf-8") as handle:
json.dump({"inflight": state.inflight, "terminal": state.terminal}, handle, sort_keys=True)
handle.write("\n")
os.chmod(name, 0o600)
os.replace(name, self.path)
finally:
if os.path.exists(name):
os.unlink(name)
class AgentPullRunner:
"""Pull, claim, execute and report one task at a time."""
def __init__(
self,
config: RunnerConfig,
transport: RunnerTransport,
state_store: JsonStateStore | None = None,
*,
random_fn: Callable[[], float] = random.random,
) -> None:
if not config.worker_id.strip():
raise RunnerError("worker_id is required")
self.config = config
self.transport = transport
self.state_store = state_store
self.random_fn = random_fn
self.state = state_store.load() if state_store else RunnerState()
def _save(self) -> None:
if self.state_store:
self.state_store.save(self.state)
def register(self) -> Mapping[str, Any]:
return self.transport.register_worker(self.config.worker_id, self.config.capabilities)
def heartbeat(self) -> Mapping[str, Any]:
return self.transport.heartbeat(self.config.worker_id)
def discovery_delay(self, failures: int) -> float:
failures = max(0, int(failures))
base = min(self.config.max_backoff_seconds, self.config.fallback_seconds * (2**failures))
jitter = self.random_fn() * max(0, self.config.jitter_seconds)
return min(self.config.max_backoff_seconds, base + jitter)
def discover_once(self) -> Mapping[str, Any] | None:
try:
task = self.transport.long_poll(self.config.worker_id, self.config.long_poll_seconds)
except Exception:
task = None
if task is None:
task = self.transport.poll(self.config.worker_id)
if not task:
return None
task_id = str(task.get("task_id", ""))
if not task_id:
raise RunnerError("task_id is required")
if task_id in self.state.terminal:
return {"status": "duplicate_suppressed", "task_id": task_id}
claimed = dict(self.transport.claim(self.config.worker_id, task))
self.state.inflight[task_id] = claimed
self._save()
return claimed
def renew(self, task_id: str) -> Mapping[str, Any]:
if task_id not in self.state.inflight:
raise RunnerError("task is not inflight")
return self.transport.renew(self.config.worker_id, task_id)
def report_progress(self, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]:
if task_id not in self.state.inflight:
raise RunnerError("task is not inflight")
return self.transport.report_progress(self.config.worker_id, task_id, payload)
def report_evidence(self, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]:
if task_id not in self.state.inflight:
raise RunnerError("task is not inflight")
return self.transport.report_evidence(self.config.worker_id, task_id, payload)
def report_result(self, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]:
if task_id in self.state.terminal:
return {"status": "duplicate_terminal_suppressed", "task_id": task_id}
if task_id not in self.state.inflight:
raise RunnerError("task is not inflight")
result = dict(self.transport.report_result(self.config.worker_id, task_id, payload))
self.state.terminal[task_id] = {"result": result, "reported_at": time.time()}
self.state.inflight.pop(task_id, None)
self._save()
return result
def recover_after_restart(self) -> list[Mapping[str, Any]]:
recovered: list[Mapping[str, Any]] = []
for task_id in tuple(self.state.inflight):
recovered.append(self.renew(task_id))
return recovered
def validate_execution(self, argv: Sequence[str], cwd: str | os.PathLike[str], env: Mapping[str, str]) -> None:
if not argv or any(not isinstance(item, str) or not item for item in argv):
raise RunnerError("argv_only_execution_required")
if any(ch in argv[0] for ch in (";", "|", "&", "`", "\n")):
raise RunnerError("shell_syntax_denied")
executable = shutil.which(argv[0])
if not executable:
raise RunnerError("executable_not_found")
if self.config.executable_allowlist and not any(
os.path.realpath(executable).startswith(os.path.realpath(prefix).rstrip(os.sep) + os.sep)
or os.path.realpath(executable) == os.path.realpath(prefix)
for prefix in self.config.executable_allowlist
):
raise RunnerError("executable_not_allowlisted")
real_cwd = os.path.realpath(os.fspath(cwd))
if self.config.working_directory_allowlist and not any(
real_cwd == os.path.realpath(prefix)
or real_cwd.startswith(os.path.realpath(prefix).rstrip(os.sep) + os.sep)
for prefix in self.config.working_directory_allowlist
):
raise RunnerError("working_directory_not_allowlisted")
if self.config.environment_allowlist and any(key not in self.config.environment_allowlist for key in env):
raise RunnerError("environment_key_not_allowlisted")
def execute_argv(self, argv: Sequence[str], cwd: str | os.PathLike[str], env: Mapping[str, str]) -> Mapping[str, Any]:
self.validate_execution(argv, cwd, env)
completed = subprocess.run(list(argv), cwd=os.fspath(cwd), env=dict(env), shell=False, text=True, capture_output=True, check=False)
return {"returncode": completed.returncode, "stdout": completed.stdout, "stderr": completed.stderr}
@@ -0,0 +1,36 @@
"""Small operational CLI for the Agent Pull Runner contract."""
from __future__ import annotations
import argparse
import json
from agent_pull_runner import AgentPullRunner, RunnerConfig
class _NoopTransport:
def register_worker(self, worker_id, capabilities): return {"status": "registered", "worker_id": worker_id, "capabilities": list(capabilities)}
def heartbeat(self, worker_id): return {"status": "alive", "worker_id": worker_id}
def long_poll(self, worker_id, timeout_seconds): return None
def poll(self, worker_id): return None
def claim(self, worker_id, task): return task
def renew(self, worker_id, task_id): return {"status": "renewed", "task_id": task_id}
def report_progress(self, worker_id, task_id, payload): return {"status": "progress_recorded", "task_id": task_id}
def report_result(self, worker_id, task_id, payload): return {"status": "result_recorded", "task_id": task_id}
def report_evidence(self, worker_id, task_id, payload): return {"status": "evidence_recorded", "task_id": task_id}
def main() -> int:
parser = argparse.ArgumentParser(description="AIOS Agent Pull Runner")
parser.add_argument("--worker-id", default="self-test-worker")
parser.add_argument("--self-test", action="store_true")
args = parser.parse_args()
runner = AgentPullRunner(RunnerConfig(worker_id=args.worker_id, fallback_seconds=60), _NoopTransport())
if args.self_test:
print(json.dumps({"status": "PASS", "registration": runner.register(), "fallback_seconds": runner.config.fallback_seconds}, sort_keys=True))
return 0
print(json.dumps({"status": "READY", "worker_id": args.worker_id}, sort_keys=True))
return 0
if __name__ == "__main__":
raise SystemExit(main())
@@ -0,0 +1,73 @@
"""HTTP adapter for Owner-authorized task writer provisioning.
Run this module instead of aios_chatgpt_actions_api.py. All existing routes are
delegated unchanged; only POST /api/task-writer/provision-authorized is added.
"""
from __future__ import annotations
import os
import urllib.parse
import aios_chatgpt_actions_api as api
from authorized_task_writer_provisioning import (
AuthorizedTaskWriterProvisioner,
OwnerContext,
ProvisioningError,
)
_original_do_post = api.H.do_POST
_ENDPOINT = "/api/task-writer/provision-authorized"
def _owner_authorization(handler):
scope = os.getenv(
"AIOS_GITOPS_OWNER_PROVISION_SCOPE", "writer:provision"
)
auth = handler.need(scope)
if not auth:
return None
if auth.get("role") != "owner":
handler.sendj({"error": "OWNER_AUTHORIZATION_REQUIRED"}, 403)
return None
identity = str(auth.get("account") or "").strip()
if not identity:
handler.sendj({"error": "OWNER_IDENTITY_REQUIRED"}, 403)
return None
return auth
def _authorized_do_post(self):
path = urllib.parse.urlparse(self.path).path.rstrip("/") or "/"
if path != _ENDPOINT:
return _original_do_post(self)
auth = _owner_authorization(self)
if not auth:
return
data = self.body()
owner_reference = str(data.get("owner_authorization_reference") or "")
try:
result = AuthorizedTaskWriterProvisioner().provision(
data,
OwnerContext(
identity=str(auth["account"]),
role=str(auth["role"]),
authorization_reference=owner_reference,
),
)
return self.sendj(result)
except ProvisioningError as exc:
body = {"error": exc.code}
if exc.details:
body["details"] = exc.details
return self.sendj(body, exc.status)
except Exception:
# Do not expose tool output, paths containing credentials, or traceback data.
return self.sendj({"error": "AUTHORIZED_WRITER_BRIDGE_INTERNAL_ERROR"}, 500)
api.H.do_POST = _authorized_do_post
if __name__ == "__main__":
api.main()
@@ -0,0 +1,89 @@
"""Governed API facade for branch-bound repair leases.
The facade is transport-neutral so the permanent control plane can expose it
through its existing authenticated loopback/API seam without creating a second
Git writer or a global credential registry.
"""
from __future__ import annotations
from typing import Any, Mapping
from branch_repair_lease import LeaseError, RepairLease, RepairLeaseManager, sanitized_lease_view
class BranchRepairLeaseAPI:
"""Validate operation envelopes and return sanitized lease metadata."""
def __init__(self, manager: RepairLeaseManager) -> None:
self.manager = manager
@staticmethod
def _required(request: Mapping[str, Any], name: str) -> Any:
value = request.get(name)
if value is None or value == "":
raise LeaseError(f"missing required field: {name}")
return value
@staticmethod
def _view(lease: RepairLease) -> dict[str, Any]:
return {"status": "ok", "lease": sanitized_lease_view(lease)}
def handle(self, request: Mapping[str, Any]) -> dict[str, Any]:
try:
operation = self._required(request, "operation")
owner = str(self._required(request, "owner_identity"))
if operation == "issue":
lease = self.manager.issue(
task_id=str(self._required(request, "task_id")),
owner_identity=owner,
branch=str(self._required(request, "branch")),
description=str(self._required(request, "description")),
allowed_paths=tuple(self._required(request, "allowed_paths")),
ttl_seconds=int(request.get("ttl_seconds", 3600)),
)
elif operation == "attach_existing_branch":
lease = self.manager.attach_existing_branch(
str(self._required(request, "lease_id")),
owner_identity=owner,
expected_head_oid=str(self._required(request, "expected_head_oid")),
)
elif operation == "authorize_commit":
lease = self.manager.authorize_commit(
str(self._required(request, "lease_id")),
owner_identity=owner,
expected_head_oid=str(self._required(request, "expected_head_oid")),
paths=tuple(self._required(request, "paths")),
content=request.get("content"),
)
elif operation == "record_commit":
lease = self.manager.record_commit(
str(self._required(request, "lease_id")),
owner_identity=owner,
expected_head_oid=str(self._required(request, "expected_head_oid")),
new_head_oid=str(self._required(request, "new_head_oid")),
paths=tuple(self._required(request, "paths")),
)
elif operation == "recover":
lease = self.manager.recover(
str(self._required(request, "lease_id")), owner_identity=owner
)
elif operation == "renew":
lease = self.manager.renew(
str(self._required(request, "lease_id")),
owner_identity=owner,
ttl_seconds=int(self._required(request, "ttl_seconds")),
)
elif operation == "complete":
lease = self.manager.complete(
str(self._required(request, "lease_id")), owner_identity=owner
)
elif operation == "revoke":
lease = self.manager.revoke(
str(self._required(request, "lease_id")), owner_identity=owner
)
else:
raise LeaseError("unsupported lease operation")
return self._view(lease)
except (LeaseError, TypeError, ValueError):
return {"status": "blocked", "error": "lease_request_rejected"}
@@ -0,0 +1,48 @@
"""Root-only CLI for the branch-bound repair lease API."""
from __future__ import annotations
import argparse
import json
import os
import sys
from pathlib import Path
from aios_branch_repair_lease_api import BranchRepairLeaseAPI
from branch_repair_lease import GitProbe, LeaseStore
from branch_repair_lease import RepairLeaseManager
def main() -> int:
parser = argparse.ArgumentParser(description="AIOS branch repair lease CLI")
parser.add_argument("--store", required=True)
parser.add_argument("--bare-repo", required=True)
parser.add_argument("--request-file")
parser.add_argument("--request-json")
args = parser.parse_args()
if hasattr(os, "geteuid") and os.geteuid() != 0:
print(json.dumps({"status": "blocked", "error": "root_required"}))
return 2
if bool(args.request_file) == bool(args.request_json):
print(json.dumps({"status": "blocked", "error": "one_request_source_required"}))
return 2
try:
raw = (
Path(args.request_file).read_text(encoding="utf-8")
if args.request_file
else str(args.request_json)
)
request = json.loads(raw)
if not isinstance(request, dict):
raise ValueError("request must be an object")
api = BranchRepairLeaseAPI(RepairLeaseManager(LeaseStore(args.store), GitProbe(args.bare_repo)))
response = api.handle(request)
print(json.dumps(response, ensure_ascii=True, sort_keys=True, separators=(",", ":")))
return 0 if response.get("status") == "ok" else 1
except Exception:
print(json.dumps({"status": "blocked", "error": "lease_cli_failed"}))
return 1
if __name__ == "__main__":
raise SystemExit(main())
+152 -4
View File
@@ -24,6 +24,15 @@ from gitea_tag_publication import (
validate_version,
)
from orchestrator.action_monitor import ActionMonitor
from silent_release_hardening import (
ApprovalKeyRing,
ApprovalRevocationStore,
CapabilityAuthorizationError,
SagaConflict,
SignedApprovalVerifier,
SignedReleasePlanApproval,
SilentReleaseActionService,
)
try:
from authoritative_git import (
AuthoritativeGitError,
@@ -60,6 +69,17 @@ OPENAPI = WS / 'docs' / 'gitops' / 'openapi-chatgpt-actions.yaml'
BARE_REPO = Path(os.getenv('AIOS_GITOPS_BARE_REPO', str(ROOT / 'runtime' / 'gitea-bin' / 'data' / 'repositories' / 'aios' / 'aios.git')))
AUTHORITATIVE_GIT_DIR = Path(os.getenv('AIOS_GITOPS_AUTHORITATIVE_GIT_DIR')) if os.getenv('AIOS_GITOPS_AUTHORITATIVE_GIT_DIR') else None
REPOSITORY_IDENTITY = os.getenv('AIOS_GITOPS_REPOSITORY_IDENTITY', '')
SILENT_RELEASE_STATE_ROOT = Path(os.getenv(
'AIOS_SILENT_RELEASE_STATE_ROOT',
'/var/lib/aios-gitops/silent-release',
))
SILENT_RELEASE_EXECUTION_ENABLED = str(os.getenv(
'AIOS_SILENT_RELEASE_EXECUTION_ENABLED', 'false'
)).lower() in {'1', 'true', 'yes', 'on'}
SILENT_RELEASE_PUBLIC_KEYS = Path(os.getenv(
'AIOS_SILENT_RELEASE_PUBLIC_KEYS',
'/etc/aios-gitops/silent-release-owner-public-keys.json',
))
monitor = ActionMonitor()
BASE_MODELS = [
@@ -68,7 +88,10 @@ BASE_MODELS = [
{'id':'base.gpt-5.5','provider':'openai','model':'gpt-5.5','display_name':'GPT-5.5','scope':'global'},
]
SECRET_RE = re.compile(r'(?i)(api[_-]?key|secret[_-]?access[_-]?key|secret|password|token)\s*[:=]\s*["\']?([A-Za-z0-9_./+=:-]{16,})')
SECRET_VAL_RE = re.compile(r'(sk-[A-Za-z0-9_-]{20,}|HPUA[A-Z0-9]{10,}|SDC[A-Za-z0-9]{20,})')
SECRET_VAL_RE = re.compile(
r'(?<![A-Za-z0-9_])'
r'(sk-[A-Za-z0-9_-]{20,}|HPUA[A-Z0-9]{10,}|SDC[A-Za-z0-9]{20,})'
)
READ_ONLY_GET_PATHS = frozenset({
'/', '/health', '/api/health', '/api/openapi.yaml',
@@ -80,11 +103,23 @@ READ_ONLY_GET_PATHS = frozenset({
'/admin/ai-evolution/api/data',
'/api/release-consistency', '/api/tag-publication-status',
'/api/release-acceptance',
'/getReleaseCandidate', '/getReleasePlan', '/getReleaseAcceptance',
'/getApprovalEnvelope', '/getReleaseSaga', '/getCapabilityStatus',
'/getRecoveryStatus',
})
SILENT_RELEASE_POST_PATHS = frozenset({
'/createReleaseCandidate',
'/createReleaseIntent',
'/submitSignedApproval',
'/submitOwnerApproval',
'/startDeployment',
})
def reject_non_read_http(handler, method: str, path: str) -> bool:
if method == 'GET':
return False
if path in SILENT_RELEASE_POST_PATHS:
return False
decision = evaluate_git_write(WS)
if decision.allowed:
return False
@@ -101,6 +136,41 @@ def sh(args, check=False):
def jload(p: Path, default):
try: return json.loads(p.read_text(encoding='utf-8'))
except Exception: return default
def _file_sha256(path: Path) -> str:
digest=hashlib.sha256()
with path.open('rb') as source:
for chunk in iter(lambda: source.read(1024 * 1024), b''):
digest.update(chunk)
return digest.hexdigest()
def _build_silent_release_service() -> SilentReleaseActionService:
verifier=None
if SILENT_RELEASE_PUBLIC_KEYS.is_file():
verifier=SignedApprovalVerifier(
ApprovalKeyRing.from_file(SILENT_RELEASE_PUBLIC_KEYS),
ApprovalRevocationStore(
SILENT_RELEASE_STATE_ROOT / 'revocations'
),
)
policy_path=WS / 'config' / 'gitops' / 'silent-release-policy.json'
policy_sha=_file_sha256(policy_path) if policy_path.is_file() else ''
return SilentReleaseActionService(
SILENT_RELEASE_STATE_ROOT,
execution_enabled=SILENT_RELEASE_EXECUTION_ENABLED,
approval_verifier=verifier,
expected_repository=REPOSITORY_IDENTITY,
policy_sha256=policy_sha,
)
# Construct once during service startup. If production execution is enabled
# without every trusted builder/orchestrator dependency, construction fails
# closed before the HTTP server accepts traffic.
_SILENT_RELEASE_SERVICE = _build_silent_release_service()
def silent_release_service() -> SilentReleaseActionService:
return _SILENT_RELEASE_SERVICE
def relpath(path: str) -> str:
x = str(path or '').replace('\\','/').strip()
if not x or x.startswith('/') or '..' in Path(x).parts or Path(x).parts[0] == '.git': raise ValueError('invalid path')
@@ -468,12 +538,51 @@ class H(BaseHTTPRequestHandler):
if path == '/api/release-acceptance':
request=publication_request_from_query(q)
return self.sendj(publication_publisher().run_release_acceptance(request))
if path in (
'/getReleaseCandidate',
'/getReleasePlan',
'/getReleaseAcceptance',
'/getApprovalEnvelope',
'/getReleaseSaga',
'/getCapabilityStatus',
):
candidate_id=(q.get('candidate_id') or [''])[0]
if not candidate_id:
return self.sendj({'error':'candidate_id is required'},400)
service=silent_release_service()
if path == '/getReleaseCandidate':
return self.sendj(
service.get_release_candidate(candidate_id)
)
if path == '/getApprovalEnvelope':
return self.sendj(
service.get_approval_envelope(candidate_id)
)
if path == '/getReleasePlan':
return self.sendj(service.get_release_plan(candidate_id))
if path == '/getReleaseAcceptance':
return self.sendj(service.get_acceptance(candidate_id))
if path == '/getCapabilityStatus':
return self.sendj(
service.get_capability_status(candidate_id)
)
return self.sendj(service.get_release_saga(candidate_id))
if path == '/getRecoveryStatus':
return self.sendj(
silent_release_service().get_recovery_status()
)
# Admin AI Evolution Dashboard
if path in ('/admin/ai-evolution','/admin/ai-evolution/'):
return self.ai_evolution_page()
if path in ('/admin/ai-evolution/api/data'):
return self.ai_evolution_data()
return self.sendj({'error':'not_found','path':path},404)
except FileNotFoundError:
return self.sendj({'error':'release candidate not found'},404)
except ValueError as e:
error=redact(e)
monitor.record_failure('do_GET', error, {'path':path})
return self.sendj({'error':error},400)
except Exception as e:
error=redact(e)
monitor.record_failure('do_GET', error, {'path':path})
@@ -489,6 +598,32 @@ class H(BaseHTTPRequestHandler):
if reject_non_read_http(self, 'POST', path): return
data=self.body()
try:
if path == '/createReleaseCandidate':
a=self.need('release:candidate')
if not a: return
return self.sendj(
silent_release_service().create_release_candidate(data)
)
if path == '/createReleaseIntent':
a=self.need('release:candidate')
if not a: return
return self.sendj(
silent_release_service().create_release_intent(data)
)
if path in ('/submitOwnerApproval','/submitSignedApproval'):
a=self.need('release:approve')
if not a: return
return self.sendj(
silent_release_service().submit_owner_approval(data)
)
if path == '/startDeployment':
a=self.need('release:deploy')
if not a: return
return self.sendj(
silent_release_service().start_deployment(
str(data.get('candidate_id') or '')
)
)
if path in ('/api/branch','/api/create-branch'): return self.branch(data)
if path in ('/api/file','/api/commit-file'): return self.file(data)
if path == '/api/commit': return self.commit(data)
@@ -499,9 +634,22 @@ class H(BaseHTTPRequestHandler):
if path == '/api/model-priority': return self.model_priority(data)
if path == '/api/nl': return self.nl(data)
return self.sendj({'error':'not_found','path':path},404)
except (ValueError, FileNotFoundError) as e:
error=redact(e)
monitor.record_failure('do_POST', error, {'path':path})
return self.sendj({'error':error},400)
except CapabilityAuthorizationError as e:
error=redact(e)
monitor.record_failure('do_POST', error, {'path':path})
return self.sendj({'error':error},403)
except SagaConflict as e:
error=redact(e)
monitor.record_failure('do_POST', error, {'path':path})
return self.sendj({'error':error},409)
except Exception as e:
monitor.record_failure('do_POST', str(e), {'path':path})
return self.sendj({'error':str(e)},500)
error=redact(e)
monitor.record_failure('do_POST', error, {'path':path})
return self.sendj({'error':error},500)
def do_PATCH(self):
path=urllib.parse.urlparse(self.path).path.rstrip('/')
if reject_non_read_http(self, 'PATCH', path): return
@@ -703,7 +851,7 @@ return{vs,cmds,ar,cmd,cdet,tot,sok,sfail,runs,baseline,refreshMsg,loadCmd,selVer
def api_spec():
return {'version':'V1R1C0-chatgpt-actions','openapi':'/api/openapi.yaml','read':['GET /api/base','GET /api/ver','GET /api/diff','GET /api/patch','GET /api/model','GET /api/run','GET /admin/ai-evolution'],'write':['POST /api/branch','POST /api/file','POST /api/commit','POST /api/patch','POST /api/tag','POST /api/rollback','POST /api/model','PUT /api/model-priority','POST /api/nl'],'policy':{'bearer_required_for_write':True,'direct_main_push':False,'delete_tag':False,'owner_approval_required_for_v_level':True,'api_keys_in_git':False,'git_secret_field':'secret_ref','online_git_sync':True}}
return {'version':'V1R1C0-chatgpt-actions','openapi':'/api/openapi.yaml','read':['GET /api/base','GET /api/ver','GET /api/diff','GET /api/patch','GET /api/model','GET /api/run','GET /getReleaseCandidate','GET /getApprovalEnvelope','GET /getReleaseSaga','GET /admin/ai-evolution'],'write':['POST /api/branch','POST /api/file','POST /api/commit','POST /api/patch','POST /api/tag','POST /api/rollback','POST /api/model','PUT /api/model-priority','POST /api/nl','POST /createReleaseCandidate','POST /submitOwnerApproval','POST /startDeployment'],'policy':{'bearer_required_for_write':True,'direct_main_push':False,'delete_tag':False,'owner_approval_required_for_v_level':True,'api_keys_in_git':False,'git_secret_field':'secret_ref','online_git_sync':True,'silent_release_execution_enabled':SILENT_RELEASE_EXECUTION_ENABLED,'silent_release_shadow_default':True}}
def main():
print(f'AIOS ChatGPT GitOps adapter listening on {HOST}:{PORT}')
@@ -0,0 +1,69 @@
"""CLI client for the governed local writer service."""
from __future__ import annotations
import argparse
import json
import os
import socket
import sys
def _parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Send one governed-local-writer request")
parser.add_argument("--socket", required=True, help="AF_UNIX socket path")
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument("--request", help="Request JSON string")
group.add_argument("--request-file", help="Path to JSON request file")
return parser.parse_args()
def _load_request_text(args: argparse.Namespace) -> str:
if args.request is not None:
return args.request
request_path = os.fspath(args.request_file)
with open(request_path, "r", encoding="utf-8") as request_file:
return request_file.read()
def send_request(socket_path: str, request_text: str) -> dict[str, object]:
"""Open one Unix socket, send one NDJSON request, and parse one NDJSON response."""
payload = json.loads(request_text)
if not isinstance(payload, dict):
raise ValueError("request must be a JSON object")
with socket.socket(socket.AF_UNIX, socket.SOCK_STREAM) as client:
client.connect(socket_path)
client.sendall((json.dumps(payload) + "\n").encode("utf-8"))
response_chunks: list[str] = []
while True:
chunk = client.recv(1024 * 1024).decode("utf-8")
if not chunk:
break
response_chunks.append(chunk)
if "\n" in chunk:
break
response_text = "".join(response_chunks)
response_line = response_text.split("\n", 1)[0].strip()
if not response_line:
raise RuntimeError("empty response")
response = json.loads(response_line)
if not isinstance(response, dict):
raise RuntimeError("invalid response")
return response
def main() -> None:
args = _parse_args()
request_text = _load_request_text(args)
response = send_request(args.socket, request_text)
print(json.dumps(response, separators=(",", ":"), sort_keys=True))
if __name__ == "__main__":
try:
main()
except Exception as exc: # pragma: no cover - CLI entrypoint behavior
print(json.dumps({"status": "error", "error": str(exc)}), file=sys.stderr)
raise SystemExit(1)
@@ -0,0 +1,931 @@
"""Unix-socket service for governed local writer operations."""
from __future__ import annotations
import json
import hashlib
import argparse
import importlib.util
import os
import re
import socket
import subprocess
import sys
import tempfile
import uuid
from dataclasses import dataclass, field
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Callable, Mapping, Protocol
from governed_local_writer import (
AppendCommitRequest,
AuthorizationError,
BranchPolicy,
CreateBranchRequest,
GovernedAuthorization,
PeerIdentity,
PRINCIPAL,
TASK_ID,
extract_peer_identity,
contains_secret_like_content,
load_authorization_file,
validate_file_changes,
)
class ExistingGitOpsHandler:
"""Delegate repository mutations to the installed GitOps handler seam."""
def __init__(
self,
*,
workspace_by_branch: Mapping[str, str | os.PathLike[str]],
bare_repo: str | os.PathLike[str],
api_source: str | os.PathLike[str],
) -> None:
self.workspace_by_branch = {str(k): Path(v) for k, v in workspace_by_branch.items()}
self.bare_repo = Path(bare_repo)
self.api_source = Path(api_source)
@staticmethod
def _restore_env(previous: Mapping[str, str | None]) -> None:
for key, value in previous.items():
if value is None:
os.environ.pop(key, None)
else:
os.environ[key] = value
def _load_api(self, workspace: Path):
if not self.api_source.is_file() or self.api_source.is_symlink():
raise RuntimeError("installed GitOps handler source is unavailable")
keys = (
"AIOS_GITOPS_WORKSPACE",
"AIOS_GITOPS_BARE_REPO",
"AIOS_GITOPS_WRITES_ENABLED",
"PYTHONDONTWRITEBYTECODE",
"GIT_CONFIG_COUNT",
"GIT_CONFIG_KEY_0",
"GIT_CONFIG_VALUE_0",
"GIT_CONFIG_KEY_1",
"GIT_CONFIG_VALUE_1",
)
previous = {key: os.environ.get(key) for key in keys}
os.environ.update(
{
"AIOS_GITOPS_WORKSPACE": str(workspace),
"AIOS_GITOPS_BARE_REPO": str(self.bare_repo),
"AIOS_GITOPS_WRITES_ENABLED": "true",
"PYTHONDONTWRITEBYTECODE": "1",
"GIT_CONFIG_COUNT": "2",
"GIT_CONFIG_KEY_0": "safe.directory",
"GIT_CONFIG_VALUE_0": str(workspace),
"GIT_CONFIG_KEY_1": "safe.directory",
"GIT_CONFIG_VALUE_1": str(self.bare_repo),
}
)
module_name = f"_governed_gitops_api_{uuid.uuid4().hex}"
try:
spec = importlib.util.spec_from_file_location(module_name, self.api_source)
if spec is None or spec.loader is None:
raise RuntimeError("unable to load installed GitOps handler")
module = importlib.util.module_from_spec(spec)
sys.modules[module_name] = module
spec.loader.exec_module(module)
decision = module.evaluate_git_write(module.WS)
if not decision.allowed:
raise RuntimeError("existing GitOps write gate denied bootstrap")
return module, previous, module_name
except Exception:
sys.modules.pop(module_name, None)
self._restore_env(previous)
raise
def _workspace(self, branch: str) -> Path:
workspace = self.workspace_by_branch.get(branch)
if workspace is None or not workspace.is_dir():
raise RuntimeError("branch workspace is not authorized")
return workspace
@staticmethod
def _git(workspace: Path, args: list[str]) -> str:
result = subprocess.run(
["git", *args], cwd=workspace, text=True, capture_output=True, check=False
)
if result.returncode != 0:
raise RuntimeError("governed handler git probe failed")
return result.stdout.strip()
def create_branch(self, request: CreateBranchRequest) -> Mapping[str, object]:
workspace = self._workspace(request.branch)
# A prior handler call may have completed the fast-forward sync before
# the process observed a post-commit error. Reconcile only an exact,
# metadata-only result; never overwrite or move an existing ref here.
try:
existing = self._git(
workspace,
[
"--git-dir",
str(self.bare_repo),
"rev-parse",
f"refs/heads/{request.branch}",
],
)
except RuntimeError:
existing = ""
if existing:
parent = self._git(workspace, ["--git-dir", str(self.bare_repo), "rev-parse", f"{existing}^1"])
changed = self._git(
workspace,
[
"--git-dir",
str(self.bare_repo),
"diff-tree",
"--no-commit-id",
"--name-only",
"-r",
existing,
],
).splitlines()
safe_branch = re.sub(r"[^A-Za-z0-9_.-]+", "-", request.branch).strip("-")
if parent != request.base_oid or changed != [f"patches/{safe_branch}.branch.json"]:
raise RuntimeError("existing branch is not an exact metadata-only bootstrap")
return {"head_oid": existing, "parent_oid": parent, "handler_status": "reconciled"}
module, previous, module_name = self._load_api(workspace)
phase = "loaded"
try:
phase = "verify_base"
parent = self._git(
workspace,
["--git-dir", str(self.bare_repo), "rev-parse", "refs/heads/main"],
)
if parent != request.base_oid:
raise RuntimeError("authorized base OID drift")
phase = "checkout"
module.checkout(request.branch, base=request.base_oid, create=True)
phase = "write_metadata"
safe_branch = re.sub(r"[^A-Za-z0-9_.-]+", "-", request.branch).strip("-")
metadata_path = f"patches/{safe_branch}.branch.json"
module.write_json(
metadata_path,
{
"schema": "aios.governed-local-writer-bootstrap.v1",
"task_id": TASK_ID,
"principal": PRINCIPAL,
"branch": request.branch,
"parent_oid": request.base_oid,
"authorization_sha256": request.authorization_digest,
},
)
phase = "commit_and_sync"
result = module.commit(
[metadata_path], "chore(governance): bootstrap repair lease branch"
)
head_oid = self._git(workspace, ["rev-parse", "HEAD"])
first_parent = self._git(workspace, ["rev-parse", "HEAD^1"])
if first_parent != request.base_oid or (request.head_oid and head_oid != request.head_oid):
raise RuntimeError("bootstrap branch lineage check failed")
return {
"head_oid": head_oid,
"parent_oid": first_parent,
"handler_status": str(result.get("status")),
}
except Exception as exc:
raise RuntimeError(f"handler_phase={phase};error={exc.__class__.__name__}") from exc
finally:
sys.modules.pop(module_name, None)
self._restore_env(previous)
def append_commit(self, request: AppendCommitRequest) -> Mapping[str, object]:
workspace = self._workspace(request.branch)
module, previous, module_name = self._load_api(workspace)
phase = "loaded"
try:
phase = "verify_head"
current = self._git(workspace, ["rev-parse", "HEAD"])
if current != request.expected_head_oid:
raise RuntimeError("branch head advanced or is stale")
phase = "checkout"
module.checkout(request.branch, base=request.expected_head_oid, create=False)
phase = "write_files"
paths: list[str] = []
for change in request.changes:
paths.append(module.write_file(change.path, change.content))
phase = "commit_and_sync"
result = module.commit(paths, request.commit_message)
head_oid = self._git(workspace, ["rev-parse", "HEAD"])
first_parent = self._git(workspace, ["rev-parse", "HEAD^1"])
if first_parent != request.expected_head_oid or head_oid == first_parent:
raise RuntimeError("append commit lineage check failed")
return {
"head_oid": head_oid,
"parent_oid": first_parent,
"handler_status": str(result.get("status")),
}
except Exception as exc:
raise RuntimeError(f"handler_phase={phase};error={exc.__class__.__name__}") from exc
finally:
sys.modules.pop(module_name, None)
self._restore_env(previous)
class GovernedWriterHandler(Protocol):
"""Small adapter interface for real writer backend handlers."""
def create_branch(self, request: CreateBranchRequest) -> Mapping[str, object]:
"""Create a branch using internal repo services."""
def append_commit(self, request: AppendCommitRequest) -> Mapping[str, object]:
"""Append a commit using internal repo services."""
class AuditSink(Protocol):
"""Pluggable sink for sanitized audit records."""
def append(self, record: "AuditRecord") -> None:
"""Persist one audit record."""
@dataclass
class AuditRecord:
"""Sanitized audit log entry."""
authorization_digest: str
peer_pid: int
peer_uid: int
peer_gid: int
peer_executable: str | None
timestamp: datetime
operation: str
branch: str
result: str
class InMemoryAuditSink:
"""Simple in-memory audit log used by tests."""
def __init__(self) -> None:
self.records: list[AuditRecord] = []
def append(self, record: AuditRecord) -> None:
self.records.append(record)
@dataclass
class _AuthorizationState:
"""Tracks one-shot operation consumption."""
create_done: set[str] = field(default_factory=set)
append_done: set[str] = field(default_factory=set)
consumed: bool = False
class GovernedLocalWriterService:
"""Policy-aware AF_UNIX service that calls an injected handler."""
def __init__(
self,
socket_path: str | os.PathLike[str],
authorization: GovernedAuthorization,
handler: GovernedWriterHandler,
*,
audit_sink: AuditSink | None = None,
peer_identity_fn: Callable[[socket.socket], PeerIdentity] | None = None,
state_path: str | os.PathLike[str] | None = None,
) -> None:
self.socket_path = Path(socket_path)
self.authorization = authorization
self.handler = handler
self.peer_identity_fn = peer_identity_fn or extract_peer_identity
self.state = _AuthorizationState()
self.audit_sink = audit_sink or InMemoryAuditSink()
self.state_path = Path(state_path) if state_path is not None else None
self._load_state()
def _load_state(self) -> None:
"""Restore one-shot state without restoring request or secret material."""
if self.state_path is None or not self.state_path.exists():
return
try:
raw = json.loads(self.state_path.read_text(encoding="utf-8"))
if not isinstance(raw, dict):
raise ValueError("state must be an object")
if raw.get("authorization_sha256") != self.authorization.authorization_sha256:
raise ValueError("state authorization digest mismatch")
self.state.create_done = {str(v) for v in raw.get("create_done", [])}
self.state.append_done = {str(v) for v in raw.get("append_done", [])}
self.state.consumed = bool(raw.get("consumed", False))
except (OSError, ValueError, TypeError, json.JSONDecodeError) as exc:
raise AuthorizationError("authorization state is invalid") from exc
def _persist_state(self) -> None:
if self.state_path is None:
return
self.state_path.parent.mkdir(parents=True, exist_ok=True)
payload = {
"authorization_sha256": self.authorization.authorization_sha256,
"create_done": sorted(self.state.create_done),
"append_done": sorted(self.state.append_done),
"consumed": self.state.consumed,
}
fd, temp_name = tempfile.mkstemp(prefix=".writer-state-", dir=self.state_path.parent)
try:
os.fchmod(fd, 0o600)
with os.fdopen(fd, "w", encoding="utf-8") as stream:
json.dump(payload, stream, sort_keys=True, separators=(",", ":"))
stream.write("\n")
os.replace(temp_name, self.state_path)
os.chmod(self.state_path, 0o600)
finally:
if os.path.exists(temp_name):
os.unlink(temp_name)
def _new_audit(self, peer: PeerIdentity, operation: str, branch: str, result: str) -> None:
self.audit_sink.append(
AuditRecord(
authorization_digest=self.authorization.authorization_sha256,
peer_pid=peer.pid,
peer_uid=peer.uid,
peer_gid=peer.gid,
peer_executable=peer.executable,
timestamp=datetime.now(tz=timezone.utc),
operation=operation,
branch=branch,
result=result,
)
)
def _require_request_field(self, request: Mapping[str, Any], field_name: str) -> str:
value = request.get(field_name)
if not isinstance(value, str) or not value:
raise AuthorizationError(f"{field_name} must be a non-empty string")
return value
def _branch_policy(self, branch: str) -> BranchPolicy:
policy = self.authorization.branch_operations.get(branch)
if policy is None:
raise AuthorizationError("branch is not authorized")
return policy
def _ensure_not_consumed(self) -> None:
if self.state.consumed:
raise AuthorizationError("authorization consumed")
if datetime.now(tz=timezone.utc) >= self.authorization.expires_at:
raise AuthorizationError("authorization expired")
def _mark_success(self, op: str, branch: str) -> None:
if op == "create_branch":
self.state.create_done.add(branch)
elif op == "append_commit":
self.state.append_done.add(branch)
if self.state.create_done.issuperset(self.authorization.create_branches) and self.state.append_done.issuperset(
self.authorization.append_branches
):
self.state.consumed = True
self._persist_state()
def _validate_request(self, request: Mapping[str, Any]) -> dict[str, Any]:
task_id = self._require_request_field(request, "task_id")
if task_id != TASK_ID:
raise AuthorizationError("invalid task_id")
principal = self._require_request_field(request, "principal")
if principal != PRINCIPAL:
raise AuthorizationError("invalid principal")
operation = self._require_request_field(request, "op")
if operation not in {"create_branch", "append_commit"}:
raise AuthorizationError("unsupported operation")
if operation in self.authorization.denied_operations:
raise AuthorizationError("operation denied by policy")
branch = self._require_request_field(request, "branch")
self._ensure_not_consumed()
policy = self._branch_policy(branch)
if operation == "create_branch":
if policy.create_branch is None:
raise AuthorizationError("branch does not allow create_branch")
if branch in self.state.create_done:
raise AuthorizationError("create_branch already used for this branch")
base_oid = self._require_request_field(request, "base_oid")
raw_head_oid = request.get("head_oid")
if raw_head_oid is not None and (not isinstance(raw_head_oid, str) or not raw_head_oid):
raise AuthorizationError("head_oid must be a non-empty string when supplied")
head_oid = str(raw_head_oid) if raw_head_oid is not None else None
if base_oid != policy.create_branch.base_oid:
raise AuthorizationError("base_oid mismatch")
if policy.create_branch.head_oid is not None and head_oid != policy.create_branch.head_oid:
raise AuthorizationError("head_oid mismatch")
return {
"operation": operation,
"branch": branch,
"base_oid": base_oid,
"head_oid": head_oid,
}
if policy.append_commit is None:
raise AuthorizationError("branch does not allow append_commit")
if policy.create_branch is not None and branch not in self.state.create_done:
raise AuthorizationError("append_commit requires prior successful create_branch")
if branch in self.state.append_done:
raise AuthorizationError("append_commit already used for this branch")
expected_head = policy.append_commit.expected_head_oid
if "expected_head_oid" in request:
supplied_head = self._require_request_field(request, "expected_head_oid")
elif "current_head_oid" in request:
supplied_head = self._require_request_field(request, "current_head_oid")
else:
raise AuthorizationError("expected_head_oid or current_head_oid required")
if not re.fullmatch(r"[0-9a-fA-F]{40}", supplied_head):
raise AuthorizationError("head check must be a full commit OID")
if expected_head is not None and supplied_head != expected_head:
raise AuthorizationError("head check failed")
if request.get("force") is True or request.get("non_fast_forward") is True:
raise AuthorizationError("non-fast-forward operations are not allowed")
if request.get("history_rewrite") is True:
raise AuthorizationError("history rewrite is not allowed")
parent_oids = request.get("parent_oids")
if parent_oids is not None:
if not isinstance(parent_oids, list) or parent_oids != [supplied_head]:
raise AuthorizationError("parent_oids must contain exactly expected_head_oid")
changes = request.get("changes")
if not isinstance(changes, list):
raise AuthorizationError("changes must be a list")
validated_changes = validate_file_changes(changes, policy.append_commit.allowed_paths)
commit_message = request.get("commit_message", "chore(governance): append governed commit")
if not isinstance(commit_message, str) or not commit_message.strip() or len(commit_message) > 200:
raise AuthorizationError("commit_message must be a short non-empty string")
if "\n" in commit_message or "\r" in commit_message or contains_secret_like_content(commit_message):
raise AuthorizationError("commit_message failed policy")
return {
"operation": operation,
"branch": branch,
"expected_head_oid": supplied_head,
"changes": validated_changes,
"commit_message": commit_message.strip(),
}
def process_request(self, raw_request: Mapping[str, Any], peer: PeerIdentity) -> Mapping[str, Any]:
"""Run one request through policy checks and handler adapter."""
operation = raw_request.get("op") if isinstance(raw_request.get("op"), str) else "unknown"
branch = raw_request.get("branch") if isinstance(raw_request.get("branch"), str) else ""
try:
details = self._validate_request(raw_request)
operation = str(details["operation"])
branch = str(details["branch"])
if operation == "create_branch":
self.handler.create_branch(
CreateBranchRequest(
branch=branch,
base_oid=str(details["base_oid"]),
head_oid=(
str(details["head_oid"])
if details.get("head_oid") is not None
else None
),
authorization_digest=self.authorization.authorization_sha256,
)
)
self._mark_success("create_branch", branch)
self._new_audit(peer, operation, branch, "ok")
return {
"status": "ok",
"operation": operation,
"branch": branch,
"result": "applied",
}
self.handler.append_commit(
AppendCommitRequest(
branch=branch,
expected_head_oid=str(details["expected_head_oid"]),
changes=tuple(details["changes"]),
authorization_digest=self.authorization.authorization_sha256,
commit_message=str(details["commit_message"]),
)
)
self._mark_success("append_commit", branch)
self._new_audit(peer, operation, branch, "ok")
return {
"status": "ok",
"operation": operation,
"branch": branch,
"result": "applied",
}
except Exception as error:
self._new_audit(peer, operation or "unknown", branch, f"error:{error.__class__.__name__}")
if isinstance(error, PermissionError):
code = "permission_denied"
elif isinstance(error, AuthorizationError):
code = "request_denied"
else:
code = "handler_failed"
if code == "handler_failed" and str(error).startswith("handler_phase="):
code = str(error)
return {
"status": "error",
"operation": operation or "unknown",
"branch": branch,
"error": code,
}
def _write_json(self, connection: socket.socket, payload: Mapping[str, Any]) -> None:
connection.sendall((json.dumps(payload) + "\n").encode("utf-8"))
def _handle_client(self, connection: socket.socket) -> None:
with connection:
try:
peer = self.peer_identity_fn(connection)
except Exception:
self._write_json(
connection,
{
"status": "error",
"operation": "connection",
"branch": "",
"error": "peer identity validation failed",
},
)
return
with connection.makefile("r", encoding="utf-8") as stream:
for raw_line in stream:
line = raw_line.strip()
if not line:
continue
try:
request = json.loads(line)
except json.JSONDecodeError:
self._write_json(
connection,
{
"status": "error",
"operation": "request",
"branch": "",
"error": "invalid json",
},
)
continue
if not isinstance(request, Mapping):
self._write_json(
connection,
{
"status": "error",
"operation": "request",
"branch": "",
"error": "request must be a JSON object",
},
)
continue
self._write_json(connection, self.process_request(request, peer))
def _bind_socket(self) -> socket.socket:
if self.socket_path.exists():
raise RuntimeError("governed writer socket already exists")
parent = self.socket_path.parent
if not parent.is_dir():
raise RuntimeError("governed writer socket parent is missing")
if parent.is_symlink():
raise RuntimeError("governed writer socket parent must not be a symlink")
server = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
server.bind(str(self.socket_path))
os.chmod(self.socket_path, 0o600)
server.listen(16)
return server
def serve(self) -> None:
"""Serve newline-delimited JSON over a filesystem AF_UNIX socket."""
server = self._bind_socket()
try:
while True:
connection, _ = server.accept()
self._handle_client(connection)
finally:
server.close()
if self.socket_path.exists():
self.socket_path.unlink()
class JsonAuditSink:
"""Append-only root-owned audit sink containing no request payloads."""
def __init__(self, path: str | os.PathLike[str]) -> None:
self.path = Path(path)
self.path.parent.mkdir(parents=True, exist_ok=True)
def append(self, record: AuditRecord) -> None:
payload = {
"authorization_sha256": record.authorization_digest,
"peer_pid": record.peer_pid,
"peer_uid": record.peer_uid,
"peer_gid": record.peer_gid,
"peer_executable": record.peer_executable,
"timestamp": record.timestamp.isoformat(),
"operation": record.operation,
"branch": record.branch,
"result": record.result,
}
line = (json.dumps(payload, sort_keys=True, separators=(",", ":")) + "\n").encode("utf-8")
fd = os.open(self.path, os.O_WRONLY | os.O_CREAT | os.O_APPEND, 0o600)
try:
os.fchmod(fd, 0o600)
os.write(fd, line)
finally:
os.close(fd)
# V2R1C6 durable receipt overlay. This subclasses the original root-only
# writer in-place so the existing AF_UNIX/SO_PEERCRED and handler policy remain
# the only mutation path.
_LegacyGovernedLocalWriterService = GovernedLocalWriterService
class GovernedLocalWriterService(_LegacyGovernedLocalWriterService):
"""Existing writer with durable request receipts and safe recovery."""
def _load_state(self) -> None:
super()._load_state()
self.state.receipts = {}
if self.state_path is None or not self.state_path.exists():
return
raw = json.loads(self.state_path.read_text(encoding="utf-8"))
if not isinstance(raw, dict):
raise AuthorizationError("authorization state is invalid")
receipts = raw.get("receipts", {})
if not isinstance(receipts, dict):
raise AuthorizationError("authorization receipts are invalid")
self.state.receipts = {
str(key): dict(value)
for key, value in receipts.items()
if isinstance(value, dict)
}
def _persist_state(self) -> None:
if self.state_path is None:
return
self.state_path.parent.mkdir(parents=True, exist_ok=True)
payload = {
"authorization_sha256": self.authorization.authorization_sha256,
"create_done": sorted(self.state.create_done),
"append_done": sorted(self.state.append_done),
"consumed": bool(self.state.consumed),
"receipts": self.state.receipts,
}
fd, temp_name = tempfile.mkstemp(prefix=".writer-state-", dir=self.state_path.parent)
try:
os.fchmod(fd, 0o600)
with os.fdopen(fd, "w", encoding="utf-8") as stream:
json.dump(payload, stream, sort_keys=True, separators=(",", ":"))
stream.write("\n")
os.replace(temp_name, self.state_path)
os.chmod(self.state_path, 0o600)
finally:
if os.path.exists(temp_name):
os.unlink(temp_name)
def _request_identity(self, request: Mapping[str, Any]) -> dict[str, str]:
canonical = json.dumps(dict(request), sort_keys=True, separators=(",", ":"), ensure_ascii=True)
fallback = hashlib.sha256(canonical.encode("utf-8")).hexdigest()
request_id = request.get("request_id")
idempotency_key = request.get("idempotency_key")
request_id = request_id.strip() if isinstance(request_id, str) and request_id.strip() else "legacy-" + fallback
idempotency_key = idempotency_key.strip() if isinstance(idempotency_key, str) and idempotency_key.strip() else "legacy-" + fallback
for field_name, value in (("request_id", request_id), ("idempotency_key", idempotency_key)):
if len(value) > 200 or "\n" in value or "\r" in value:
raise AuthorizationError(field_name + " is invalid")
return {
"request_id": request_id,
"idempotency_key": idempotency_key,
"request_fingerprint": fallback,
}
def _find_receipt(self, details: Mapping[str, Any], compare_fingerprint: bool = True):
request_id = str(details["request_id"])
idempotency_key = str(details["idempotency_key"])
fingerprint = str(details.get("request_fingerprint", ""))
for receipt in self.state.receipts.values():
if receipt.get("request_id") == request_id or receipt.get("idempotency_key") == idempotency_key:
if compare_fingerprint and receipt.get("request_fingerprint") != fingerprint:
raise AuthorizationError("idempotency key reused with different request")
return receipt
return None
@staticmethod
def _receipt_response(receipt: Mapping[str, object], status: str = "receipt") -> dict[str, object]:
response = {
"status": status,
"request_id": receipt.get("request_id"),
"idempotency_key": receipt.get("idempotency_key"),
"operation": receipt.get("operation"),
"branch": receipt.get("branch", ""),
"receipt_status": receipt.get("status"),
}
for field_name in ("commit_oid", "final_head", "parent_oid", "audit_sha256", "error_code", "safe_to_retry"):
if field_name in receipt:
response[field_name] = receipt[field_name]
return response
def _new_receipt(self, details: Mapping[str, Any]) -> dict[str, object]:
return {
"request_id": str(details["request_id"]),
"idempotency_key": str(details["idempotency_key"]),
"request_fingerprint": str(details["request_fingerprint"]),
"operation": str(details["operation"]),
"branch": str(details.get("branch", "")),
"expected_head_oid": str(details.get("expected_head_oid", "")),
"status": "accepted",
"created_at": datetime.now(tz=timezone.utc).isoformat(),
"safe_to_retry": False,
}
def _new_audit(self, peer: PeerIdentity, operation: str, branch: str, result: str) -> str:
super()._new_audit(peer, operation, branch, result)
payload = {
"authorization_sha256": self.authorization.authorization_sha256,
"peer_pid": peer.pid,
"peer_uid": peer.uid,
"peer_gid": peer.gid,
"peer_executable": peer.executable,
"operation": operation,
"branch": branch,
"result": result,
}
return hashlib.sha256(json.dumps(payload, sort_keys=True, separators=(",", ":")).encode("utf-8")).hexdigest()
def _mark_success(self, op: str, branch: str) -> None:
if op == "create_branch":
self.state.create_done.add(branch)
elif op in {"append_commit", "append_files", "append_batch"}:
self.state.append_done.add(branch)
# The provisioning grant is one-shot; the task writer lease itself must
# support multiple append receipts for the exact authorized branches.
self.state.consumed = False
self._persist_state()
def _validate_durable(self, raw_request: Mapping[str, Any]) -> dict[str, Any]:
operation = self._require_request_field(raw_request, "op")
identity = self._request_identity(raw_request)
if operation == "query_receipt":
return {"operation": operation, **identity}
if operation == "recover_request":
branch = self._require_request_field(raw_request, "branch")
policy = self._branch_policy(branch)
if policy.append_commit is None:
raise AuthorizationError("branch does not allow recovery")
expected = raw_request.get("expected_head_oid") or policy.append_commit.expected_head_oid
if not isinstance(expected, str) or not re.fullmatch(r"[0-9a-fA-F]{40}", expected):
raise AuthorizationError("recovery requires expected_head_oid")
return {"operation": operation, "branch": branch, "expected_head_oid": expected.lower(), **identity}
if operation not in {"create_branch", "append_commit", "append_files", "append_batch"}:
raise AuthorizationError("unsupported operation")
request = dict(raw_request)
if operation in {"append_files", "append_batch"}:
request["op"] = "append_commit"
if operation in {"append_commit", "append_files", "append_batch"}:
self.state.consumed = False
branch = request.get("branch")
removed = False
dynamic_expected = False
if isinstance(branch, str) and branch in self.state.append_done:
self.state.append_done.remove(branch)
removed = True
policy = self.authorization.branch_operations.get(branch)
if policy is not None and policy.append_commit is not None and policy.append_commit.expected_head_oid is not None:
from governed_local_writer import BranchAppendPolicy, BranchPolicy
self.authorization.branch_operations[branch] = BranchPolicy(create_branch=policy.create_branch, append_commit=BranchAppendPolicy(branch=branch, expected_head_oid=None, allowed_paths=policy.append_commit.allowed_paths))
dynamic_expected = True
try:
details = super()._validate_request(request)
finally:
if dynamic_expected:
policy = self.authorization.branch_operations.get(str(branch))
if policy is not None and policy.append_commit is not None:
from governed_local_writer import BranchAppendPolicy, BranchPolicy
self.authorization.branch_operations[str(branch)] = BranchPolicy(create_branch=policy.create_branch, append_commit=BranchAppendPolicy(branch=str(branch), expected_head_oid=str(raw_request.get("expected_head_oid") or ""), allowed_paths=policy.append_commit.allowed_paths))
if removed:
self.state.append_done.add(str(branch))
details.update(identity)
details["operation"] = operation
return details
def _recover_backend(self, details: Mapping[str, Any]) -> dict[str, object]:
if not hasattr(self.handler, "_workspace") or not hasattr(self.handler, "_git"):
raise RuntimeError("recovery backend unavailable")
workspace = self.handler._workspace(str(details["branch"]))
expected = str(details["expected_head_oid"])
current = self.handler._git(workspace, ["rev-parse", "HEAD"])
if current == expected:
return {"status": "not_started", "final_head": current}
parent = self.handler._git(workspace, ["rev-parse", "HEAD^1"])
if parent != expected:
raise RuntimeError("external branch advance during recovery")
return {"status": "applied", "commit_oid": current, "final_head": current, "parent_oid": parent}
def process_request(self, raw_request: Mapping[str, Any], peer: PeerIdentity) -> Mapping[str, Any]:
operation = raw_request.get("op") if isinstance(raw_request.get("op"), str) else "unknown"
branch = raw_request.get("branch") if isinstance(raw_request.get("branch"), str) else ""
try:
details = self._validate_durable(raw_request)
operation = str(details["operation"])
branch = str(details.get("branch", ""))
if operation == "query_receipt":
receipt = self._find_receipt(details, compare_fingerprint=False)
return self._receipt_response(receipt) if receipt is not None else {"status": "not_found", "request_id": details["request_id"], "idempotency_key": details["idempotency_key"]}
if operation == "recover_request":
receipt = self._find_receipt(details, compare_fingerprint=False)
if receipt is None:
return {"status": "not_found", "request_id": details["request_id"], "idempotency_key": details["idempotency_key"], "safe_to_retry": True}
if receipt.get("status") == "applied":
return self._receipt_response(receipt)
recovered = self._recover_backend(details)
if recovered.get("status") == "applied":
receipt.update({"status": "applied", "commit_oid": recovered.get("commit_oid"), "final_head": recovered.get("final_head"), "parent_oid": recovered.get("parent_oid"), "safe_to_retry": False})
self._mark_success("append_commit", branch)
return self._receipt_response(receipt)
receipt["safe_to_retry"] = True
self._persist_state()
return self._receipt_response(receipt)
receipt = self._find_receipt(details)
if receipt is not None:
return self._receipt_response(receipt)
receipt = self._new_receipt(details)
self.state.receipts[str(details["request_id"])] = receipt
self._persist_state()
if operation == "create_branch":
result = self.handler.create_branch(CreateBranchRequest(branch=branch, base_oid=str(details["base_oid"]), head_oid=str(details["head_oid"]) if details.get("head_oid") is not None else None, authorization_digest=self.authorization.authorization_sha256))
commit_oid = str(result.get("head_oid", ""))
parent_oid = str(result.get("parent_oid", ""))
audit_sha = self._new_audit(peer, operation, branch, "ok")
receipt.update({"status": "applied", "commit_oid": commit_oid, "final_head": commit_oid, "parent_oid": parent_oid, "audit_sha256": audit_sha, "safe_to_retry": False})
self._mark_success("create_branch", branch)
return self._receipt_response(receipt, status="ok") | {"result": "applied"}
result = self.handler.append_commit(AppendCommitRequest(branch=branch, expected_head_oid=str(details["expected_head_oid"]), changes=tuple(details["changes"]), authorization_digest=self.authorization.authorization_sha256, commit_message=str(details["commit_message"])))
commit_oid = str(result.get("head_oid", ""))
parent_oid = str(result.get("parent_oid", ""))
audit_sha = self._new_audit(peer, operation, branch, "ok")
receipt.update({"status": "applied", "commit_oid": commit_oid, "final_head": commit_oid, "parent_oid": parent_oid, "audit_sha256": audit_sha, "safe_to_retry": False})
self._mark_success("append_commit", branch)
return self._receipt_response(receipt, status="ok") | {"result": "applied"}
except Exception as error:
try:
identity = self._request_identity(raw_request)
receipt = self.state.receipts.get(identity["request_id"])
if receipt is not None and receipt.get("status") == "accepted":
receipt.update({"status": "failed", "error_code": error.__class__.__name__, "safe_to_retry": False})
self._persist_state()
except Exception:
pass
self._new_audit(peer, operation, branch, "error:" + error.__class__.__name__)
return {"status": "error", "operation": operation, "branch": branch, "error": "request_denied" if isinstance(error, AuthorizationError) else "handler_failed"}
def _parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="AIOS governed local writer")
parser.add_argument("--socket", required=True)
parser.add_argument("--authorization", required=True)
parser.add_argument("--state", required=True)
parser.add_argument("--audit", required=True)
parser.add_argument("--repair-branch", required=True)
parser.add_argument("--repair-workspace", required=True)
parser.add_argument("--candidate-branch", required=True)
parser.add_argument("--candidate-workspace", required=True)
parser.add_argument("--bare-repo", required=True)
parser.add_argument("--api-source", required=True)
return parser.parse_args()
def main() -> int:
if hasattr(os, "geteuid") and os.geteuid() != 0:
print("governed local writer requires uid 0", file=sys.stderr)
return 2
args = _parse_args()
authorization = load_authorization_file(args.authorization)
handler = ExistingGitOpsHandler(
workspace_by_branch={
args.repair_branch: args.repair_workspace,
args.candidate_branch: args.candidate_workspace,
},
bare_repo=args.bare_repo,
api_source=args.api_source,
)
service = GovernedLocalWriterService(
args.socket,
authorization,
handler,
audit_sink=JsonAuditSink(args.audit),
state_path=args.state,
)
service.serve()
return 0
if __name__ == "__main__": # pragma: no cover - systemd entrypoint
raise SystemExit(main())
@@ -0,0 +1,316 @@
"""Task-scoped hardened wrapper around the existing Action API."""
from __future__ import annotations
import json
import os
import subprocess
from pathlib import Path
import urllib.parse
import aios_chatgpt_actions_api as api
import aios_isolated_writer_policy as policy
from isolated_writer_authz import AuthorizationError, authorize
from isolated_writer_commit_file import AtomicCommitFile, CommitFileError, OperationLedger
from isolated_writer_token_registry import TokenRegistry
_reject = api.reject_non_read_http
_write_file = api.write_file
_sync_branch = api.sync_branch
_original_do_get = api.H.do_GET
api.OPENAPI = Path(os.getenv("AIOS_GITOPS_ISOLATED_WRITER_OPENAPI", str(api.OPENAPI)))
def _registry() -> TokenRegistry:
return TokenRegistry(Path(os.environ["AIOS_GITOPS_TOKEN_REGISTRY_V2"]))
def _atomic_writer() -> AtomicCommitFile:
return AtomicCommitFile(
workspace=api.WS,
bare_repo=api.BARE_REPO,
ledger=OperationLedger(Path(os.environ["AIOS_GITOPS_WRITER_OPERATION_LEDGER"])),
remote=os.getenv("AIOS_GITOPS_GITEA_PUSH_REMOTE", ""),
audit=policy.append_audit_record,
)
def _bearer(handler) -> str:
value = handler.headers.get("Authorization", "")
return value[7:].strip() if value.startswith("Bearer ") else ""
def _authorize(handler, *, method: str, path: str, branch=None, target_path=None):
grant = authorize(
_registry(),
_bearer(handler),
method=method,
path=path,
source_ip=str(handler.client_address[0]),
workspace=api.WS,
branch=branch,
target_path=target_path,
)
grant.record["_source_ip"] = str(handler.client_address[0])
return grant
def _authorization_error(handler, exc: AuthorizationError):
return handler.sendj({"error": exc.code}, exc.status)
def _isolated_runtime_status(handler):
branch = api.sh(["rev-parse", "--abbrev-ref", "HEAD"], True)["stdout"]
head = api.sh(["rev-parse", "HEAD"], True)["stdout"]
return handler.sendj({
"service": os.getenv(
"AIOS_GITOPS_SERVICE_AUDIENCE",
"aios-isolated-writer.service",
),
"status": "active",
"runtime_role": os.getenv("AIOS_GITOPS_RUNTIME_ROLE"),
"workspace_role": os.getenv("AIOS_GITOPS_WORKSPACE_ROLE"),
"writes_enabled": os.getenv("AIOS_GITOPS_WRITES_ENABLED", "").lower()
in {"1", "true", "yes", "on"},
"allowed_branch": policy.allowed_branch(),
"approved_base_oid": policy.allowed_base_oid(),
"workspace_branch": branch,
"workspace_head_oid": head,
"repository_identity": os.getenv("AIOS_GITOPS_REPOSITORY_IDENTITY"),
"token_policy_version": 2,
"atomic_commit_file": True,
"path_policy_sha256": policy.configured_path_policy_sha256(),
})
def _isolated_do_get(self):
path = urllib.parse.urlparse(self.path).path.rstrip("/") or "/"
if path in {"/health", "/api/health"}:
return self.sendj({
"status": "ok",
"service": os.getenv(
"AIOS_GITOPS_SERVICE_AUDIENCE",
"aios-isolated-writer.service",
),
})
if path not in {"/api/runtime-status", "/api/baseline", "/api/patches"}:
return self.sendj({"error": "isolated_writer_operation_forbidden"}, 403)
try:
_authorize(self, method="GET", path=path)
except AuthorizationError as exc:
return _authorization_error(self, exc)
if path == "/api/runtime-status":
return _isolated_runtime_status(self)
return _original_do_get(self)
def _read_commit_file_body(handler):
try:
length = int(handler.headers.get("Content-Length", "0"))
except ValueError:
raise CommitFileError("writer_content_length_rejected") from None
if length <= 0:
raise CommitFileError("writer_request_object_required")
if length > 1_100_000:
raise CommitFileError("writer_request_too_large", 413)
raw = handler.rfile.read(length)
try:
value = json.loads(raw.decode("utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError):
raise CommitFileError("writer_request_json_rejected") from None
if not isinstance(value, dict):
raise CommitFileError("writer_request_object_required")
return value
def _isolated_do_post(self):
path = urllib.parse.urlparse(self.path).path.rstrip("/") or "/"
if path != "/api/commit-file":
return self.sendj({"error": "isolated_writer_operation_forbidden"}, 403)
try:
data = _read_commit_file_body(self)
branch = str(data.get("branch") or "")
target_path = str(data.get("path") or "")
grant = _authorize(
self,
method="POST",
path=path,
branch=branch,
target_path=target_path,
)
if api.content_has_secret(str(data.get("content") or "")):
raise CommitFileError("writer_secret_content_rejected", 403)
result = _atomic_writer().execute(grant, data)
return self.sendj(result)
except AuthorizationError as exc:
return _authorization_error(self, exc)
except CommitFileError as exc:
return self.sendj({"error": exc.code}, exc.status)
except PermissionError:
return self.sendj({"error": "isolated_writer_policy_rejected"}, 403)
def reject_non_read_http(handler, method: str, path: str) -> bool:
if _reject(handler, method, path):
return True
try:
policy.validate_endpoint(method, path, api.WS, api.BARE_REPO)
return False
except PermissionError as exc:
handler.sendj({"error": str(exc)}, 403)
return True
def checkout(branch, base="main", create=False):
api.ensure_git()
target = api.branch_name(branch)
policy.validate_branch_request({"branch": target, "base": base})
exists = api.sh(["rev-parse", "--verify", target])["code"] == 0
if exists:
if api.sh(["rev-parse", "--abbrev-ref", "HEAD"], True)["stdout"] != target:
api.sh(["checkout", target], True)
elif create:
api.sh(["checkout", "-b", target, base], True)
else:
raise PermissionError("isolated_writer_branch_missing")
policy.require_service_invariants(api.WS, api.BARE_REPO)
return target
def write_file(path, content):
policy.validate_path(path, api.WS)
return _write_file(path, content)
def sync_branch(branch: str):
policy.validate_sync_branch(branch)
local = _sync_branch(branch)
remote = os.getenv("AIOS_GITOPS_GITEA_PUSH_REMOTE", "").strip()
if remote != "gitea-http":
raise PermissionError("isolated_writer_gitea_remote_rejected")
head = api.sh(["rev-parse", "HEAD"], True)["stdout"]
pushed = subprocess.run(
["git", "push", "--porcelain", remote, f"HEAD:refs/heads/{branch}"],
cwd=api.WS,
text=True,
capture_output=True,
)
if pushed.returncode != 0:
raise RuntimeError("isolated_writer_gitea_push_failed")
verified = subprocess.run(
["git", "ls-remote", "--heads", remote, f"refs/heads/{branch}"],
cwd=api.WS,
text=True,
capture_output=True,
)
if verified.returncode != 0:
raise RuntimeError("isolated_writer_gitea_verification_failed")
fields = verified.stdout.strip().split()
if len(fields) != 2 or fields[0] != head or fields[1] != f"refs/heads/{branch}":
raise RuntimeError("isolated_writer_gitea_oid_mismatch")
return {
"status": "verified",
"target": "authoritative-bare+gitea",
"branch": branch,
"oid": head,
"local": local.get("status"),
}
def commit(paths, msg):
policy.validate_commit_paths(paths, api.WS)
api.ensure_git()
selected = [api.relpath(p) for p in paths] if paths else ["."]
api.sh(["add", *selected], True)
policy.validate_staged_paths(api.WS)
branch = api.sh(["rev-parse", "--abbrev-ref", "HEAD"], True)["stdout"]
if not api.sh(["status", "--short"])["stdout"]:
return {"status": "no_changes", "branch": branch, "sync": sync_branch(branch)}
out = api.sh(["commit", "-m", msg], True)
branch = api.sh(["rev-parse", "--abbrev-ref", "HEAD"], True)["stdout"]
return {
"status": "committed",
"branch": branch,
"rev": api.sh(["rev-parse", "--short", "HEAD"], True)["stdout"],
"git": out,
"sync": sync_branch(branch),
}
def audit(actor, action, target, payload):
policy.append_audit_record(
{
"at": api.now(),
"actor": actor,
"action": action,
"target": target,
"payload": api.sanitize(payload),
}
)
def _validate_request(self, data, validate):
try:
validate(data)
return True
except PermissionError as exc:
self.sendj({"error": str(exc)}, 403)
return False
def _checked_branch(self, data):
if not _validate_request(self, data, policy.validate_branch_request):
return
return _branch(self, data)
def _checked_file(self, data):
data = dict(data)
data.setdefault("branch", policy.allowed_branch())
if not _validate_request(self, data, policy.validate_branch_request):
return
if not _validate_request(
self,
{"path": data.get("path", "")},
lambda item: policy.validate_path(item["path"], api.WS),
):
return
return _file(self, data)
def _checked_commit(self, data):
data = dict(data)
data.setdefault("branch", policy.allowed_branch())
if not _validate_request(self, data, policy.validate_branch_request):
return
return _commit(self, data)
def _checked_patch(self, data):
data = dict(data)
data["branch"] = policy.allowed_branch()
if not _validate_request(self, data, policy.validate_branch_request):
return
return _patch(self, data)
_branch, _file, _commit, _patch = api.H.branch, api.H.file, api.H.commit, api.H.patch
api.reject_non_read_http = reject_non_read_http
api.checkout, api.write_file, api.sync_branch, api.commit, api.audit = (
checkout,
write_file,
sync_branch,
commit,
audit,
)
api.H.branch, api.H.file, api.H.commit, api.H.patch = (
_checked_branch,
_checked_file,
_checked_commit,
_checked_patch,
)
api.H.do_GET = _isolated_do_get
api.H.do_POST = _isolated_do_post
if __name__ == "__main__":
policy.require_service_invariants(api.WS, api.BARE_REPO)
api.ThreadingHTTPServer((api.HOST, api.PORT), api.H).serve_forever()
@@ -0,0 +1,310 @@
"""Fail-closed policy used only by the task-scoped Action writer."""
from __future__ import annotations
import os
import json
import stat
import subprocess
import re
from pathlib import Path, PurePosixPath, PureWindowsPath
from urllib.parse import unquote
import isolated_writer_path_policy as path_policy
WRITE_ENDPOINTS = frozenset({"/api/commit-file"})
ALLOWED_ROOTS = (
"active-runtime",
"docs",
"patches",
"reports",
"scripts",
"test",
"tests",
)
DENIED_ROOTS = ("baseline", "config", "deploy", "deployments", "ops", "runtime", "services")
_TRUE = frozenset({"1", "true", "yes", "on"})
_FALSE = frozenset({"0", "false", "no", "off"})
_DENY_FLAGS = (
"AIOS_GITOPS_ALLOW_MAIN_WRITE",
"AIOS_GITOPS_ALLOW_TAG_WRITE",
"AIOS_GITOPS_ALLOW_BASELINE_WRITE",
"AIOS_GITOPS_ALLOW_REGISTRY_WRITE",
"AIOS_GITOPS_ALLOW_ROLLBACK",
"AIOS_GITOPS_ALLOW_DEPLOYMENT",
)
def enforcement_required() -> bool:
return os.getenv("AIOS_GITOPS_ISOLATED_WRITER_ENFORCEMENT", "") == "required"
def allowed_branch() -> str:
value = os.getenv("AIOS_GITOPS_ALLOWED_BRANCH", "").strip()
if not value:
raise PermissionError("isolated_writer_missing_allowed_branch")
return value
def allowed_base_oid() -> str:
value = os.getenv("AIOS_GITOPS_ALLOWED_BASE_OID", "").strip().lower()
if len(value) != 40 or any(ch not in "0123456789abcdef" for ch in value):
raise PermissionError("isolated_writer_invalid_allowed_base_oid")
return value
def _git(workspace: Path, *args: str) -> str:
result = subprocess.run(["git", *args], cwd=workspace, text=True, capture_output=True)
if result.returncode != 0:
raise PermissionError("isolated_writer_git_validation_failed")
return result.stdout.strip()
def require_service_invariants(workspace: Path, bare_repo: Path) -> None:
if not enforcement_required():
return
if os.getenv("AIOS_GITOPS_WRITES_ENABLED", "").strip().lower() not in _TRUE:
raise PermissionError("isolated_writer_write_gate_required")
if os.getenv("AIOS_GITOPS_RUNTIME_ROLE", "") != "isolated-writer":
raise PermissionError("isolated_writer_role_required")
if os.getenv("AIOS_GITOPS_WORKSPACE_ROLE", "") != "isolated-writer":
raise PermissionError("isolated_writer_workspace_role_required")
if os.getenv("AIOS_GITOPS_TOKEN_POLICY_VERSION_REQUIRED", "") != "2":
raise PermissionError("isolated_writer_token_policy_v2_required")
if not os.getenv("AIOS_GITOPS_SERVICE_AUDIENCE", "").strip():
raise PermissionError("isolated_writer_service_audience_required")
if not os.getenv("AIOS_GITOPS_TOKEN_REGISTRY_V2", "").strip():
raise PermissionError("isolated_writer_token_registry_required")
if not os.getenv("AIOS_GITOPS_WRITER_OPERATION_LEDGER", "").strip():
raise PermissionError("isolated_writer_operation_ledger_required")
path_policy.configured_path_policy()
for name in _DENY_FLAGS:
if os.getenv(name, "").strip().lower() not in _FALSE:
raise PermissionError("isolated_writer_forbidden_capability_flag")
if os.getenv("AIOS_GITOPS_REPOSITORY_IDENTITY", "") != "aios/aios":
raise PermissionError("isolated_writer_repository_identity_rejected")
configured_bare_value = os.getenv("AIOS_GITOPS_AUTHORITATIVE_GIT_DIR", "").strip()
if not configured_bare_value:
raise PermissionError("isolated_writer_authoritative_repo_rejected")
configured_bare = Path(configured_bare_value)
if configured_bare.resolve(strict=True) != bare_repo.resolve(strict=True):
raise PermissionError("isolated_writer_authoritative_repo_rejected")
writer_root_value = os.getenv("AIOS_GITOPS_APPROVED_WRITER_ROOT", "").strip()
if not writer_root_value:
raise PermissionError("isolated_writer_workspace_outside_approved_root")
writer_root = Path(writer_root_value).resolve(strict=True)
lexical_workspace = Path(os.path.abspath(str(workspace)))
canonical_workspace = lexical_workspace.resolve(strict=True)
if lexical_workspace != canonical_workspace:
raise PermissionError("isolated_writer_workspace_symlink_rejected")
try:
canonical_workspace.relative_to(writer_root)
except (ValueError, FileNotFoundError):
raise PermissionError("isolated_writer_workspace_outside_approved_root") from None
if _git(workspace, "rev-parse", "--abbrev-ref", "HEAD") != allowed_branch():
raise PermissionError("isolated_writer_current_branch_rejected")
approved_base = allowed_base_oid()
try:
_git(workspace, "cat-file", "-e", f"{approved_base}^{{commit}}")
subprocess.run(
["git", "merge-base", "--is-ancestor", approved_base, "HEAD"],
cwd=workspace,
check=True,
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
except (PermissionError, subprocess.CalledProcessError):
raise PermissionError("isolated_writer_approved_base_not_ancestor") from None
if _git(workspace, "merge-base", approved_base, "HEAD") != approved_base:
raise PermissionError("isolated_writer_merge_base_rejected")
common = Path(_git(workspace, "rev-parse", "--git-common-dir"))
if not common.is_absolute():
common = (workspace / common).resolve()
if common.resolve() != bare_repo.resolve():
raise PermissionError("isolated_writer_common_repo_rejected")
def approved_audit_log_path() -> Path:
root_value = os.getenv("AIOS_GITOPS_APPROVED_AUDIT_ROOT", "").strip()
log_value = os.getenv("AIOS_GITOPS_AUDIT_LOG", "").strip()
if not root_value or not log_value:
raise PermissionError("isolated_writer_audit_path_missing")
lexical_root = Path(os.path.abspath(root_value))
lexical_log = Path(os.path.abspath(log_value))
root = lexical_root.resolve(strict=True)
if lexical_root != root or root.is_symlink() or not root.is_dir():
raise PermissionError("isolated_writer_audit_root_rejected")
try:
lexical_log.relative_to(root)
except ValueError:
raise PermissionError("isolated_writer_audit_path_outside_approved_root") from None
if lexical_log.parent.resolve(strict=True) != lexical_log.parent:
raise PermissionError("isolated_writer_audit_parent_symlink_rejected")
if lexical_log.exists() and lexical_log.is_symlink():
raise PermissionError("isolated_writer_audit_log_symlink_rejected")
os.chmod(root, 0o750)
os.chmod(lexical_log.parent, 0o750)
return lexical_log
def append_audit_record(record: dict) -> None:
path = approved_audit_log_path()
flags = os.O_WRONLY | os.O_APPEND | os.O_CREAT
if hasattr(os, "O_NOFOLLOW"):
flags |= os.O_NOFOLLOW
descriptor = os.open(path, flags, 0o600)
try:
info = os.fstat(descriptor)
if not stat.S_ISREG(info.st_mode):
raise PermissionError("isolated_writer_audit_log_not_regular")
os.fchmod(descriptor, 0o600)
line = json.dumps(
_audit_sanitize(record),
ensure_ascii=False,
sort_keys=True,
separators=(",", ":"),
).encode("utf-8") + b"\n"
os.write(descriptor, line)
os.fsync(descriptor)
finally:
os.close(descriptor)
parent_fd = os.open(path.parent, os.O_RDONLY | getattr(os, "O_DIRECTORY", 0))
try:
os.fsync(parent_fd)
finally:
os.close(parent_fd)
def _audit_sanitize(value):
if isinstance(value, dict):
return {
key: (
"[REDACTED]"
if any(
marker in str(key).lower()
for marker in (
"authorization",
"credential",
"password",
"secret",
"token",
"private_key",
)
)
and str(key).lower() != "token_id"
else _audit_sanitize(inner)
)
for key, inner in value.items()
}
if isinstance(value, list):
return [_audit_sanitize(item) for item in value]
if isinstance(value, str) and re.search(
r"(?i)(authorization\s*:\s*bearer|(?:token|password|secret)\s*[:=])",
value,
):
return "[REDACTED]"
return value
def validate_endpoint(method: str, path: str, workspace: Path, bare_repo: Path) -> None:
if not enforcement_required():
return
if method != "POST" or path not in WRITE_ENDPOINTS:
raise PermissionError("isolated_writer_operation_forbidden")
require_service_invariants(workspace, bare_repo)
def validate_branch_request(data: dict) -> None:
if not enforcement_required():
return
branch = allowed_branch() if "branch" not in data else str(data.get("branch") or "").strip()
if branch != allowed_branch():
raise PermissionError("isolated_writer_branch_rejected")
if str(data.get("base", "main")).strip() != "main":
raise PermissionError("isolated_writer_base_rejected")
if data.get("force") or data.get("force_push"):
raise PermissionError("isolated_writer_force_rejected")
def validate_path(path: str, workspace: Path | None = None) -> str:
if not isinstance(path, str):
raise PermissionError("isolated_writer_path_rejected")
value = path
if not value or value != value.strip() or "\x00" in value or "\\" in value:
raise PermissionError("isolated_writer_path_rejected")
try:
decoded = unquote(value, errors="strict")
except (UnicodeDecodeError, ValueError):
raise PermissionError("isolated_writer_path_rejected") from None
if decoded != value:
raise PermissionError("isolated_writer_path_rejected")
windows = PureWindowsPath(value)
posix = PurePosixPath(value)
parts = value.split("/")
if (
posix.is_absolute()
or windows.is_absolute()
or bool(windows.drive)
or any(part in {"", ".", ".."} for part in parts)
):
raise PermissionError("isolated_writer_path_rejected")
first = parts[0]
service_exact_path = False
if first in DENIED_ROOTS:
try:
path_policy.require_service_path("commit_file.atomic", value)
service_exact_path = True
except path_policy.PathPolicyError as exc:
raise PermissionError(str(exc)) from None
if value != "README.md" and first not in ALLOWED_ROOTS and not service_exact_path:
raise PermissionError("isolated_writer_path_outside_allowed_roots")
if workspace is not None:
lexical_root = Path(os.path.abspath(str(workspace)))
root = lexical_root.resolve(strict=True)
if lexical_root != root:
raise PermissionError("isolated_writer_workspace_symlink_rejected")
current = root
for part in parts:
current = current / part
if current.is_symlink():
raise PermissionError("isolated_writer_path_symlink_rejected")
candidate = (root / value).resolve(strict=False)
try:
candidate.relative_to(root)
except ValueError:
raise PermissionError("isolated_writer_path_escapes_workspace") from None
return value
def validate_content_for_path(path: str, content: str) -> None:
if path.split("/", 1)[0] not in DENIED_ROOTS:
return
try:
path_policy.validate_content("commit_file.atomic", path, content)
except path_policy.PathPolicyError as exc:
raise PermissionError(str(exc)) from None
def configured_path_policy_sha256() -> str | None:
try:
return path_policy.configured_path_policy_sha256()
except path_policy.PathPolicyError as exc:
raise PermissionError(str(exc)) from None
def validate_commit_paths(paths: list[str], workspace: Path | None = None) -> None:
if enforcement_required():
for path in paths:
if str(path).strip() != ".":
validate_path(str(path), workspace)
def validate_staged_paths(workspace: Path) -> None:
if not enforcement_required():
return
for path in _git(workspace, "diff", "--cached", "--name-only").splitlines():
validate_path(path, workspace)
def validate_sync_branch(branch: str) -> None:
if enforcement_required() and branch != allowed_branch():
raise PermissionError("isolated_writer_sync_branch_rejected")
@@ -0,0 +1,144 @@
"""HTTP adapter for audited Owner authorization and writer provisioning.
This module layers on top of ``aios_authorized_task_writer_api`` so existing
routes and the legacy exact ``writer:provision`` route remain available. The
new endpoint does not mint or modify bearer credentials. It authenticates an
existing, non-wildcard, explicitly allow-listed Owner credential and then
invokes the audited broker.
Endpoint:
POST /api/governance/owner-authorizations
"""
from __future__ import annotations
import os
import urllib.parse
from typing import Any, Mapping
import aios_authorized_task_writer_api as bridge_api
from authorized_task_writer_provisioning import ProvisioningError
from owner_authorization_broker import (
OwnerAuthorizationBroker,
OwnerAuthorizationError,
OwnerPrincipal,
)
api = bridge_api.api
_original_do_post = api.H.do_POST
_ENDPOINT = "/api/governance/owner-authorizations"
def _enabled() -> bool:
return os.getenv(
"AIOS_GITOPS_OWNER_AUTHORIZATION_BROKER_ENABLED", ""
).strip().lower() in {"1", "true", "yes", "on"}
def _configured_intake_scope() -> str:
scopes = [
item.strip()
for item in os.getenv(
"AIOS_GITOPS_OWNER_AUTHORIZATION_INTAKE_SCOPES", ""
).split(",")
if item.strip()
]
if len(scopes) != 1:
raise OwnerAuthorizationError(
"EXACTLY_ONE_HTTP_INTAKE_SCOPE_REQUIRED", 503
)
scope = scopes[0]
if scope == "writer:provision" or scope == "*" or scope.endswith(":*"):
raise OwnerAuthorizationError("HTTP_INTAKE_SCOPE_REJECTED", 503)
return scope
def _scope_set(auth: Mapping[str, Any]) -> frozenset[str]:
raw = auth.get("scopes")
if raw is None:
raw = auth.get("scope")
if raw is None:
return frozenset()
if isinstance(raw, str):
candidates = raw.replace(",", " ").split()
elif isinstance(raw, (list, tuple, set, frozenset)):
candidates = [str(item) for item in raw]
else:
raise OwnerAuthorizationError("OWNER_SCOPE_EVIDENCE_INVALID", 403)
return frozenset(item.strip() for item in candidates if item.strip())
def _credential_id(auth: Mapping[str, Any]) -> str:
for name in (
"credential_id",
"token_id",
"key_id",
"auth_id",
):
value = str(auth.get(name) or "").strip()
if value:
return value
return ""
def _authenticate_existing_owner(handler) -> OwnerPrincipal | None:
scope = _configured_intake_scope()
auth = handler.need(scope)
if not auth:
return None
if auth.get("role") != "owner":
handler.sendj({"error": "OWNER_ROLE_REQUIRED"}, 403)
return None
identity = str(auth.get("account") or auth.get("identity") or "").strip()
if not identity:
handler.sendj({"error": "OWNER_IDENTITY_REQUIRED"}, 403)
return None
scopes = _scope_set(auth)
if any(item == "*" or item.endswith(":*") for item in scopes):
handler.sendj(
{"error": "WILDCARD_OWNER_CREDENTIAL_REJECTED"}, 403
)
return None
return OwnerPrincipal(
identity=identity,
role="owner",
authenticated_scope=scope,
token_scopes=scopes,
credential_id=_credential_id(auth),
)
def _broker_do_post(self):
path = urllib.parse.urlparse(self.path).path.rstrip("/") or "/"
if path != _ENDPOINT:
return _original_do_post(self)
if not _enabled():
return self.sendj(
{"error": "OWNER_AUTHORIZATION_BROKER_DISABLED"}, 503
)
try:
principal = _authenticate_existing_owner(self)
if principal is None:
return
data = self.body()
result = OwnerAuthorizationBroker().authorize_and_provision(
data, principal
)
return self.sendj(result)
except (OwnerAuthorizationError, ProvisioningError) as exc:
body = {"error": exc.code}
if exc.details:
body["details"] = exc.details
return self.sendj(body, exc.status)
except Exception:
# Never expose registry data, bearer tokens, signer output or traceback.
return self.sendj(
{"error": "OWNER_AUTHORIZATION_BROKER_INTERNAL_ERROR"}, 500
)
api.H.do_POST = _broker_do_post
if __name__ == "__main__":
api.main()
+160
View File
@@ -0,0 +1,160 @@
#!/usr/bin/env python3
"""Loopback-only HTTP API for the task identity overlay.
Endpoints:
GET /healthz
POST /api/governance/owner-authorizations
The API authenticates only the short-lived overlay credential. It does not read
or modify the global bearer-token registry.
"""
from __future__ import annotations
import json
import os
import urllib.parse
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from typing import Any
from authorized_task_writer_provisioning import ProvisioningError
from owner_authorization_broker import OwnerAuthorizationError
from task_identity_overlay import (
MAX_BODY_BYTES,
IdentityOverlayConfig,
IdentityOverlayError,
TaskIdentityOverlayService,
)
_ENDPOINT = "/api/governance/owner-authorizations"
def _enabled() -> bool:
return os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_ENABLED", ""
).strip().lower() in {"1", "true", "yes", "on"}
def _bearer(header: str) -> str:
prefix = "Bearer "
if not header.startswith(prefix):
raise IdentityOverlayError("OVERLAY_AUTHENTICATION_REQUIRED", 401)
token = header[len(prefix):].strip()
if not token or len(token) > 4096:
raise IdentityOverlayError("OVERLAY_AUTHENTICATION_REQUIRED", 401)
return token
class OverlayHandler(BaseHTTPRequestHandler):
server_version = "aios-task-identity-overlay/1"
sys_version = ""
@property
def overlay(self) -> TaskIdentityOverlayService:
return self.server.overlay # type: ignore[attr-defined]
def log_message(self, format: str, *args: Any) -> None:
# Never log Authorization or request bodies.
return
def _send(self, status: int, value: dict[str, Any]) -> None:
payload = (
json.dumps(
value,
ensure_ascii=False,
sort_keys=True,
separators=(",", ":"),
)
+ "\n"
).encode("utf-8")
self.send_response(status)
self.send_header("Content-Type", "application/json")
self.send_header("Cache-Control", "no-store")
self.send_header("Content-Length", str(len(payload)))
self.end_headers()
self.wfile.write(payload)
def _body(self) -> dict[str, Any]:
try:
length = int(self.headers.get("Content-Length", "0"))
except ValueError:
raise IdentityOverlayError("OVERLAY_CONTENT_LENGTH_INVALID") from None
if length <= 0 or length > MAX_BODY_BYTES:
raise IdentityOverlayError("OVERLAY_BODY_SIZE_REJECTED", 413)
raw = self.rfile.read(length)
try:
value = json.loads(raw)
except (UnicodeDecodeError, json.JSONDecodeError):
raise IdentityOverlayError("OVERLAY_BODY_JSON_INVALID") from None
if not isinstance(value, dict):
raise IdentityOverlayError("OVERLAY_BODY_OBJECT_REQUIRED")
return value
def do_GET(self) -> None:
path = urllib.parse.urlparse(self.path).path.rstrip("/") or "/"
if path != "/healthz":
self._send(404, {"error": "NOT_FOUND"})
return
self._send(
200,
{
"service": "aios-task-identity-overlay",
"status": "active" if _enabled() else "disabled",
"global_token_registry_used": False,
"loopback_only": True,
},
)
def do_POST(self) -> None:
path = urllib.parse.urlparse(self.path).path.rstrip("/") or "/"
if path != _ENDPOINT:
self._send(404, {"error": "NOT_FOUND"})
return
if not _enabled():
self._send(503, {"error": "IDENTITY_OVERLAY_DISABLED"})
return
try:
token = _bearer(self.headers.get("Authorization", ""))
result = self.overlay.provision(self._body(), token)
self._send(200, result)
except (IdentityOverlayError, OwnerAuthorizationError, ProvisioningError) as exc:
body: dict[str, Any] = {"error": exc.code}
if getattr(exc, "details", None):
body["details"] = exc.details
self._send(exc.status, body)
except Exception:
self._send(500, {"error": "IDENTITY_OVERLAY_API_INTERNAL_ERROR"})
class OverlayServer(ThreadingHTTPServer):
allow_reuse_address = True
def __init__(
self,
address: tuple[str, int],
overlay: TaskIdentityOverlayService,
) -> None:
super().__init__(address, OverlayHandler)
self.overlay = overlay
def main() -> None:
host = os.getenv("AIOS_GITOPS_IDENTITY_OVERLAY_HOST", "127.0.0.1")
if host not in {"127.0.0.1", "::1", "localhost"}:
raise SystemExit("identity overlay must listen on loopback")
try:
port = int(os.getenv("AIOS_GITOPS_IDENTITY_OVERLAY_PORT", "19864"))
except ValueError:
raise SystemExit("invalid overlay port") from None
if not 1024 <= port <= 65535:
raise SystemExit("invalid overlay port")
config = IdentityOverlayConfig.from_environment()
server = OverlayServer(
(host, port),
TaskIdentityOverlayService(config),
)
server.serve_forever(poll_interval=0.25)
if __name__ == "__main__":
main()
+92
View File
@@ -0,0 +1,92 @@
#!/usr/bin/env python3
"""Root-governed CLI for issuing and revoking task identity overlays.
The CLI never prints the bearer token. It returns only the root-owned secret
reference and non-secret metadata.
"""
from __future__ import annotations
import argparse
import json
import os
import sys
from pathlib import Path
from task_identity_overlay import (
IdentityOverlayConfig,
IdentityOverlayError,
TaskIdentityOverlayStore,
)
def _write_result(path: str, value: dict) -> None:
data = json.dumps(value, sort_keys=True, separators=(",", ":")) + "\n"
if path == "-":
sys.stdout.write(data)
return
target = Path(path)
target.parent.mkdir(parents=True, exist_ok=True)
target.write_text(data, encoding="utf-8")
os.chmod(target, 0o600)
def parser() -> argparse.ArgumentParser:
value = argparse.ArgumentParser(
description="Issue/status/revoke a task-scoped identity overlay"
)
sub = value.add_subparsers(dest="command", required=True)
issue = sub.add_parser("issue")
issue.add_argument("--ttl-seconds", type=int, default=900)
issue.add_argument("--authorization-assertion-ref", required=True)
issue.add_argument(
"--operator-identity",
default=os.getenv("SUDO_USER") or os.getenv("USER") or "root",
)
issue.add_argument("--result-json", default="-")
status = sub.add_parser("status")
status.add_argument("--overlay-id", required=True)
status.add_argument("--result-json", default="-")
revoke = sub.add_parser("revoke")
revoke.add_argument("--overlay-id", required=True)
revoke.add_argument("--reason", required=True)
revoke.add_argument("--result-json", default="-")
return value
def main() -> int:
args = parser().parse_args()
try:
store = TaskIdentityOverlayStore(
IdentityOverlayConfig.from_environment()
)
if args.command == "issue":
result = store.issue(
ttl_seconds=args.ttl_seconds,
authorization_assertion_ref=args.authorization_assertion_ref,
operator_identity=args.operator_identity,
)
elif args.command == "status":
result = store.status(args.overlay_id)
else:
result = store.revoke(args.overlay_id, reason=args.reason)
_write_result(args.result_json, result)
return 0
except IdentityOverlayError as exc:
body = {"status": "BLOCKED", "error": exc.code}
if exc.details:
body["details"] = exc.details
_write_result(getattr(args, "result_json", "-"), body)
return 2
except Exception:
_write_result(
getattr(args, "result_json", "-"),
{"status": "FAILED", "error": "IDENTITY_OVERLAY_CLI_INTERNAL_ERROR"},
)
return 1
if __name__ == "__main__":
raise SystemExit(main())
@@ -0,0 +1,953 @@
"""Owner-authorized, fail-closed task writer provisioning bridge.
This module converts an authenticated Owner authorization into a canonical,
short-lived, one-time grant request. It delegates signing, validation and
worktree/writer provisioning to existing root-governed executables.
It never writes Git refs directly and never authorizes main, tag, baseline or
production mutations.
"""
from __future__ import annotations
import fcntl
import hashlib
import json
import os
import re
import secrets
import shutil
import stat
import subprocess
import tempfile
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Mapping
from signer_contract_compatibility import (
deployed_signer_help,
run_compatible_signer,
run_compatible_validator,
)
OID_RE = re.compile(r"^[0-9a-f]{40}$")
SHA256_RE = re.compile(r"^[0-9a-f]{64}$")
TASK_RE = re.compile(r"^[A-Z0-9][A-Z0-9._:-]{7,127}$")
BRANCH_RE = re.compile(r"^agent/[A-Za-z0-9][A-Za-z0-9._/-]{2,179}$")
OWNER_REF_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:/@+-]{3,255}$")
REQUIRED_REQUEST_FIELDS = frozenset(
{
"task_id",
"repository",
"base_ref",
"base_oid",
"target_branch",
"owner_authorization_reference",
"one_time_use",
"expires_in_seconds",
"production_change_authorized",
"tag_change_authorized",
"baseline_change_authorized",
"main_write_authorized",
"allowed_operations",
"denied_operations",
}
)
REQUIRED_ALLOWED_OPERATIONS = frozenset(
{
"create_worktree",
"create_target_branch_if_absent",
"commit_to_target_branch",
"push_target_branch",
}
)
REQUIRED_DENIED_OPERATIONS = frozenset(
{
"direct_main_push",
"force_push",
"tag_create",
"tag_update",
"tag_delete",
"branch_delete",
"repository_admin",
"baseline_write",
"production_deploy",
"production_marker_write",
"history_rewrite",
}
)
REQUIRED_NEGATIVE_TESTS = frozenset(
{
"main_write_denied",
"other_branch_write_denied",
"tag_create_denied",
"tag_update_denied",
"baseline_write_denied",
"force_push_denied",
"branch_delete_denied",
}
)
SENSITIVE_KEY_MARKERS = (
"authorization",
"credential",
"private_key",
"secret",
"signature",
"token",
)
class ProvisioningError(RuntimeError):
"""Stable fail-closed error returned to the API adapter."""
def __init__(
self,
code: str,
status: int = 400,
*,
details: Mapping[str, Any] | None = None,
) -> None:
super().__init__(code)
self.code = code
self.status = status
self.details = dict(details or {})
@dataclass(frozen=True)
class OwnerContext:
identity: str
role: str
authorization_reference: str
@dataclass(frozen=True)
class ProvisioningConfig:
repository_identity: str
authority_git_dir: Path
state_root: Path
signer: Path
validator: Path
provisioner: Path
governed_use_sudo: bool = False
sudo_path: Path = Path("/usr/bin/sudo")
max_expiry_seconds: int = 1800
@classmethod
def from_environment(cls) -> "ProvisioningConfig":
return cls(
repository_identity=os.getenv(
"AIOS_GITOPS_REPOSITORY_IDENTITY", "aios/aios"
).strip(),
authority_git_dir=Path(
os.getenv("AIOS_GITOPS_AUTHORITY_GIT_DIR", "/data/aios-gitops/repo.git")
),
state_root=Path(
os.getenv(
"AIOS_GITOPS_AUTHORIZED_WRITER_STATE_ROOT",
"/var/lib/aios-gitops/authorized-writer-bridge",
)
),
signer=Path(
os.getenv(
"AIOS_GITOPS_GRANT_SIGNER_ADAPTER",
"/usr/local/libexec/aios-grant-signer-adapter",
)
),
validator=Path(
os.getenv(
"AIOS_GITOPS_GRANT_VALIDATOR",
"/usr/local/libexec/aios-grant-validator",
)
),
provisioner=Path(
os.getenv(
"AIOS_GITOPS_TASK_WRITER_PROVISIONER",
"/usr/local/libexec/aios-task-writer-provisioner",
)
),
governed_use_sudo=os.getenv(
"AIOS_GITOPS_GOVERNED_USE_SUDO", ""
).strip().lower() in {"1", "true", "yes", "on"},
sudo_path=Path(
os.getenv("AIOS_GITOPS_SUDO_PATH", "/usr/bin/sudo")
),
)
def utc_now() -> datetime:
return datetime.now(timezone.utc)
def iso_utc(value: datetime) -> str:
return value.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
def canonical_json_bytes(value: Any) -> bytes:
return (
json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+ "\n"
).encode("utf-8")
def sha256_json(value: Any) -> str:
return hashlib.sha256(canonical_json_bytes(value)).hexdigest()
def _atomic_write_json(path: Path, value: Any, mode: int = 0o600) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
if path.parent.is_symlink() or path.is_symlink():
raise ProvisioningError("STATE_PATH_SYMLINK_REJECTED", 500)
fd, temporary = tempfile.mkstemp(prefix=".tmp-", dir=path.parent)
try:
os.fchmod(fd, mode)
with os.fdopen(fd, "wb", closefd=True) as stream:
stream.write(canonical_json_bytes(value))
stream.flush()
os.fsync(stream.fileno())
os.replace(temporary, path)
os.chmod(path, mode)
parent_fd = os.open(path.parent, os.O_RDONLY | getattr(os, "O_DIRECTORY", 0))
try:
os.fsync(parent_fd)
finally:
os.close(parent_fd)
finally:
if os.path.exists(temporary):
os.unlink(temporary)
def _read_json(path: Path) -> dict[str, Any]:
if path.is_symlink() or not path.is_file():
raise ProvisioningError("RESULT_FILE_INVALID", 500)
if path.stat().st_size <= 0 or path.stat().st_size > 1_048_576:
raise ProvisioningError("RESULT_FILE_INVALID", 500)
try:
value = json.loads(path.read_text(encoding="utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError):
raise ProvisioningError("RESULT_JSON_INVALID", 500) from None
if not isinstance(value, dict):
raise ProvisioningError("RESULT_OBJECT_REQUIRED", 500)
return value
def _safe_receipt_id(task_id: str, repository: str, branch: str) -> str:
return hashlib.sha256(f"{task_id}\0{repository}\0{branch}".encode()).hexdigest()
def _redacted(value: Any) -> Any:
if isinstance(value, dict):
cleaned = {}
for key, inner in value.items():
lowered = str(key).lower()
if any(marker in lowered for marker in SENSITIVE_KEY_MARKERS):
if lowered in {
"owner_authorization_reference",
"signing_key_identity",
"signer_identity",
}:
cleaned[key] = inner
else:
cleaned[key] = "[REDACTED]"
else:
cleaned[key] = _redacted(inner)
return cleaned
if isinstance(value, list):
return [_redacted(item) for item in value]
return value
def _run(
argv: list[str],
*,
env: Mapping[str, str] | None = None,
timeout: int = 60,
allowed_returncodes: tuple[int, ...] = (0,),
) -> subprocess.CompletedProcess[str]:
result = subprocess.run(
argv,
text=True,
capture_output=True,
check=False,
timeout=timeout,
env=dict(env) if env is not None else None,
)
if result.returncode not in allowed_returncodes:
raise ProvisioningError(
"GOVERNED_TOOL_FAILED",
503,
details={
"tool": Path(argv[0]).name,
"returncode": result.returncode,
"stderr": result.stderr[-1000:],
},
)
return result
def _git(git_dir: Path, *args: str, allowed_returncodes: tuple[int, ...] = (0,)) -> str:
result = _run(
["git", "--git-dir", str(git_dir), *args],
timeout=30,
allowed_returncodes=allowed_returncodes,
)
return result.stdout.strip()
def _verify_executable(path: Path, name: str) -> None:
if not path.is_absolute() or path.is_symlink() or not path.is_file():
raise ProvisioningError(f"{name}_EXECUTABLE_INVALID", 503)
info = path.stat(follow_symlinks=False)
if not stat.S_ISREG(info.st_mode) or not os.access(path, os.X_OK):
raise ProvisioningError(f"{name}_EXECUTABLE_INVALID", 503)
if info.st_mode & (stat.S_IWGRP | stat.S_IWOTH):
raise ProvisioningError(f"{name}_EXECUTABLE_PERMISSIONS_REJECTED", 503)
def _tool_argv(
config: ProvisioningConfig,
path: Path,
*args: str,
) -> list[str]:
if config.governed_use_sudo:
return [str(config.sudo_path), "-n", "--", str(path), *args]
return [str(path), *args]
def _verify_tool_contract(
path: Path,
name: str,
config: ProvisioningConfig,
) -> None:
_verify_executable(path, name)
if config.governed_use_sudo:
_verify_executable(config.sudo_path, "SUDO")
help_result = _run(_tool_argv(config, path, "--help"), timeout=15)
if name == "SIGNER":
sign_help = _run(
_tool_argv(config, path, "sign", "--help"),
timeout=15,
)
if deployed_signer_help(sign_help.stdout, sign_help.stderr):
_run(_tool_argv(config, path, "self-test"), timeout=30)
return
for flag in ("--grant-dir", "--result-json"):
if flag not in help_result.stdout and flag not in help_result.stderr:
raise ProvisioningError(f"{name}_CONTRACT_UNSUPPORTED", 503)
_run(_tool_argv(config, path, "--self-test"), timeout=30)
def _validate_request(
raw: Mapping[str, Any],
*,
owner: OwnerContext,
config: ProvisioningConfig,
) -> dict[str, Any]:
if owner.role != "owner" or not owner.identity.strip():
raise ProvisioningError("OWNER_AUTHORIZATION_REQUIRED", 403)
if set(raw) != REQUIRED_REQUEST_FIELDS:
raise ProvisioningError("REQUEST_FIELDS_REJECTED")
if raw["repository"] != config.repository_identity:
raise ProvisioningError("REPOSITORY_IDENTITY_MISMATCH", 409)
if raw["base_ref"] != "refs/heads/main":
raise ProvisioningError("BASE_REF_REJECTED", 409)
if not isinstance(raw["base_oid"], str) or not OID_RE.fullmatch(raw["base_oid"]):
raise ProvisioningError("BASE_OID_REJECTED")
if (
not isinstance(raw["target_branch"], str)
or not BRANCH_RE.fullmatch(raw["target_branch"])
or raw["target_branch"] in {"main", "master"}
):
raise ProvisioningError("TARGET_BRANCH_REJECTED")
if not isinstance(raw["task_id"], str) or not TASK_RE.fullmatch(raw["task_id"]):
raise ProvisioningError("TASK_ID_REJECTED")
if (
not isinstance(raw["owner_authorization_reference"], str)
or not OWNER_REF_RE.fullmatch(raw["owner_authorization_reference"])
or raw["owner_authorization_reference"] != owner.authorization_reference
):
raise ProvisioningError("OWNER_AUTHORIZATION_REFERENCE_MISMATCH", 403)
if raw["one_time_use"] is not True:
raise ProvisioningError("ONE_TIME_USE_REQUIRED")
expiry = raw["expires_in_seconds"]
if (
not isinstance(expiry, int)
or isinstance(expiry, bool)
or not 60 <= expiry <= config.max_expiry_seconds
):
raise ProvisioningError("GRANT_LIFETIME_REJECTED")
for name in (
"production_change_authorized",
"tag_change_authorized",
"baseline_change_authorized",
"main_write_authorized",
):
if raw[name] is not False:
raise ProvisioningError("PRODUCTION_CAPABILITY_REJECTED", 403)
allowed = raw["allowed_operations"]
denied = raw["denied_operations"]
if not isinstance(allowed, list) or frozenset(allowed) != REQUIRED_ALLOWED_OPERATIONS:
raise ProvisioningError("ALLOWED_OPERATIONS_MISMATCH", 403)
if len(allowed) != len(set(allowed)):
raise ProvisioningError("ALLOWED_OPERATIONS_MISMATCH", 403)
if not isinstance(denied, list) or frozenset(denied) != REQUIRED_DENIED_OPERATIONS:
raise ProvisioningError("DENIED_OPERATIONS_MISMATCH", 403)
if len(denied) != len(set(denied)):
raise ProvisioningError("DENIED_OPERATIONS_MISMATCH", 403)
return dict(raw)
def _prepare_state_root(path: Path) -> None:
if not path.is_absolute() or path.is_symlink():
raise ProvisioningError("STATE_ROOT_INVALID", 500)
path.mkdir(parents=True, exist_ok=True, mode=0o700)
os.chmod(path, 0o700)
if path.stat().st_mode & (stat.S_IWGRP | stat.S_IWOTH):
raise ProvisioningError("STATE_ROOT_PERMISSIONS_REJECTED", 500)
(path / "receipts").mkdir(mode=0o700, exist_ok=True)
(path / "locks").mkdir(mode=0o700, exist_ok=True)
def _prestart_git_check(
request: Mapping[str, Any],
config: ProvisioningConfig,
*,
receipt_exists: bool,
) -> dict[str, Any]:
git_dir = config.authority_git_dir
if not git_dir.is_absolute() or git_dir.is_symlink() or not git_dir.is_dir():
raise ProvisioningError("AUTHORITY_GIT_DIR_INVALID", 503)
actual_main = _git(git_dir, "rev-parse", request["base_ref"])
if actual_main != request["base_oid"]:
raise ProvisioningError(
"AUTHORIZED_BASE_OID_DRIFT",
409,
details={"expected": request["base_oid"], "actual": actual_main},
)
ref = f"refs/heads/{request['target_branch']}"
existing = _git(
git_dir,
"show-ref",
"--verify",
"--hash",
ref,
allowed_returncodes=(0, 1, 128),
)
if existing and not receipt_exists:
raise ProvisioningError(
"EXISTING_BRANCH_OID_CONFLICT",
409,
details={"branch_oid": existing},
)
if existing and receipt_exists:
ancestor = _run(
[
"git",
"--git-dir",
str(git_dir),
"merge-base",
"--is-ancestor",
request["base_oid"],
ref,
],
allowed_returncodes=(0, 1),
)
if ancestor.returncode != 0:
raise ProvisioningError("EXISTING_BRANCH_OID_CONFLICT", 409)
return {
"authoritative_main_oid": actual_main,
"target_branch_preexisting": bool(existing),
"target_branch_oid": existing or None,
}
def _build_plan(
request: Mapping[str, Any],
owner: OwnerContext,
*,
now: datetime,
) -> tuple[dict[str, Any], dict[str, Any]]:
expires_at = now + timedelta(seconds=int(request["expires_in_seconds"]))
grant_request = {
"schema": "aios.task-scoped-writer-grant-request.v1",
"task_id": request["task_id"],
"repository": request["repository"],
"base": {"ref": request["base_ref"], "oid": request["base_oid"]},
"target": {"branch": request["target_branch"]},
"execution_contract": {
"branch_creation": "governed_one_shot_provisioner",
"writer_route": "POST /api/commit-file",
"createBranch_route_required": False,
},
"owner_authorization": {
"identity": owner.identity,
"reference": owner.authorization_reference,
"authenticated_role": owner.role,
},
"allowed_operations": sorted(request["allowed_operations"]),
"denied_operations": sorted(request["denied_operations"]),
"one_time_use": True,
"authorization": {
"production_change_authorized": False,
"tag_change_authorized": False,
"baseline_change_authorized": False,
"main_write_authorized": False,
},
"issued_at": iso_utc(now),
"not_before": iso_utc(now - timedelta(seconds=2)),
"expires_at": iso_utc(expires_at),
"nonce": secrets.token_hex(32),
}
binding_digest = sha256_json(grant_request)
plan = {
"schema": "aios.task-scoped-writer-approval-plan.v1",
"approval_plan_id": f"sha256:{binding_digest}",
"grant_request_sha256": binding_digest,
"owner_authorization_reference": owner.authorization_reference,
"owner_identity": owner.identity,
"created_at": iso_utc(now),
"expires_at": iso_utc(expires_at),
"execute": True,
"production_change_authorized": False,
}
return plan, grant_request
def _expect_text(result: Mapping[str, Any], name: str) -> str:
value = result.get(name)
if not isinstance(value, str) or not value.strip():
raise ProvisioningError(f"{name.upper()}_MISSING", 503)
return value
def _verify_signer_result(
result: Mapping[str, Any],
*,
approval_plan_id: str,
) -> dict[str, str]:
if result.get("status") not in {"SIGNED", "PASS", "COMPLETED"}:
raise ProvisioningError("GRANT_SIGNING_FAILED", 503)
if result.get("approval_plan_id") != approval_plan_id:
raise ProvisioningError("APPROVAL_PLAN_ID_MISMATCH", 503)
digest = _expect_text(result, "grant_digest_sha256")
if not SHA256_RE.fullmatch(digest):
raise ProvisioningError("GRANT_DIGEST_INVALID", 503)
signer = result.get("signing_key_identity") or result.get("signer_identity")
if not isinstance(signer, str) or not signer.strip():
raise ProvisioningError("SIGNING_KEY_IDENTITY_MISSING", 503)
signature_present = bool(result.get("signature_present")) or bool(
result.get("owner_signature")
)
if not signature_present:
raise ProvisioningError("OWNER_SIGNATURE_MISSING", 503)
return {
"grant_id": _expect_text(result, "grant_id"),
"grant_digest_sha256": digest,
"signing_key_identity": signer,
}
def _verify_validator_result(
result: Mapping[str, Any],
*,
approval_plan_id: str,
grant_digest: str,
) -> None:
if result.get("status") not in {"PASS", "VALID", "COMPLETED"}:
raise ProvisioningError("GOVERNED_GRANT_VALIDATION_FAILED", 503)
if result.get("approval_plan_id") != approval_plan_id:
raise ProvisioningError("VALIDATED_PLAN_ID_MISMATCH", 503)
if result.get("grant_digest_sha256") != grant_digest:
raise ProvisioningError("VALIDATED_GRANT_DIGEST_MISMATCH", 503)
checks = result.get("checks")
required = {
"schema_valid",
"signature_valid",
"signer_trusted",
"approval_plan_active",
"repository_match",
"base_ref_match",
"base_oid_match",
"target_branch_match",
"one_time_use",
"allowed_operations_match",
"denied_operations_match",
}
if not isinstance(checks, dict) or any(checks.get(name) is not True for name in required):
raise ProvisioningError("GOVERNED_GRANT_VALIDATION_FAILED", 503)
def _verify_provisioner_result(
result: Mapping[str, Any],
request: Mapping[str, Any],
*,
grant_id: str,
validator_config_path: Path,
) -> dict[str, Any]:
if result.get("status") not in {"READY_FOR_CODE_AUTHORING", "COMPLETED"}:
raise ProvisioningError("TASK_WRITER_PROVISIONING_FAILED", 503)
# The legacy bridge created the target ref itself. Keep accepting it for
# controlled test doubles, but prefer the deployed direct one-shot writer.
if result.get("status") == "COMPLETED" and result.get("writer_mode") == "direct_one_shot_detached":
if result.get("target_branch") != request["target_branch"]:
raise ProvisioningError("PROVISIONED_BRANCH_MISMATCH", 503)
if result.get("base_oid") != request["base_oid"]:
raise ProvisioningError("PROVISIONED_BASE_MISMATCH", 503)
if result.get("writer_created") is not True:
raise ProvisioningError("TASK_WRITER_PROVISIONING_FAILED", 503)
writer_config_name = result.get("writer_config")
if not isinstance(writer_config_name, str) or not Path(writer_config_name).is_absolute():
raise ProvisioningError("WRITER_CONFIG_INVALID", 503)
writer_config_path = Path(writer_config_name)
writer_config = _read_json(writer_config_path)
if writer_config.get("status") != "READY":
raise ProvisioningError("WRITER_CONFIG_NOT_READY", 503)
if writer_config.get("target_branch") != request["target_branch"]:
raise ProvisioningError("WRITER_CONFIG_BRANCH_MISMATCH", 503)
if writer_config.get("base_oid") != request["base_oid"]:
raise ProvisioningError("WRITER_CONFIG_BASE_MISMATCH", 503)
if writer_config.get("worktree_mode") != "detached":
raise ProvisioningError("WRITER_MODE_UNSUPPORTED", 503)
workspace = writer_config.get("worktree")
if not isinstance(workspace, str) or not Path(workspace).is_absolute():
raise ProvisioningError("PROVISIONED_WORKSPACE_INVALID", 503)
credential_name = writer_config.get("runtime_credential_file")
if not isinstance(credential_name, str) or not Path(credential_name).is_absolute():
raise ProvisioningError("WRITER_CREDENTIAL_REF_INVALID", 503)
credential = Path(credential_name)
if not credential.is_file() or credential.is_symlink():
raise ProvisioningError("WRITER_CREDENTIAL_MISSING", 503)
credential_stat = credential.stat()
if credential_stat.st_uid != 0 or stat.S_IMODE(credential_stat.st_mode) != 0o400:
raise ProvisioningError("WRITER_CREDENTIAL_PERMISSIONS_INVALID", 503)
dynamic_config = _read_json(validator_config_path)
ledger_name = dynamic_config.get("consumed_grants_ledger")
if not isinstance(ledger_name, str) or not Path(ledger_name).is_absolute():
raise ProvisioningError("GRANT_LEDGER_REF_INVALID", 503)
ledger = Path(ledger_name)
consumed = False
if ledger.is_file() and not ledger.is_symlink():
for line in ledger.read_text(encoding="utf-8").splitlines():
try:
row = json.loads(line)
except json.JSONDecodeError:
continue
if (
isinstance(row, dict)
and row.get("grant_id") == grant_id
and row.get("target_branch") == request["target_branch"]
and row.get("base_oid") == request["base_oid"]
):
consumed = True
break
if not consumed:
raise ProvisioningError("ONE_TIME_GRANT_NOT_CONSUMED", 503)
scope = {
"workspace_role": "task-scoped-isolated-writer",
"writes_enabled": True,
"allowed_branch": request["target_branch"],
"allowed_base_oid": request["base_oid"],
"allow_main_write": False,
"allow_tag_write": False,
"allow_baseline_write": False,
"allow_force_push": False,
"allowed_routes": ["POST /api/commit-file"],
"writer_mode": "direct_one_shot_detached",
}
negatives = {name: True for name in sorted(REQUIRED_NEGATIVE_TESTS)}
return {
"branch": request["target_branch"],
"branch_oid": request["base_oid"],
"base_oid": request["base_oid"],
"workspace_path": workspace,
"writer_scope": scope,
"negative_tests": negatives,
"grant_consumed": True,
"remote_ref_status": "target_branch_absent_until_writer_push",
"remote_oid": None,
"detached_writer": True,
"writer_config_path": str(writer_config_path),
}
if result.get("branch") != request["target_branch"]:
raise ProvisioningError("PROVISIONED_BRANCH_MISMATCH", 503)
if result.get("base_oid") != request["base_oid"]:
raise ProvisioningError("PROVISIONED_BASE_MISMATCH", 503)
if result.get("branch_oid") != request["base_oid"]:
raise ProvisioningError("PROVISIONED_BRANCH_OID_MISMATCH", 503)
if result.get("grant_consumed") is not True:
raise ProvisioningError("ONE_TIME_GRANT_NOT_CONSUMED", 503)
workspace = result.get("workspace_path")
if not isinstance(workspace, str) or not Path(workspace).is_absolute():
raise ProvisioningError("PROVISIONED_WORKSPACE_INVALID", 503)
scope = result.get("writer_scope")
if not isinstance(scope, dict):
raise ProvisioningError("WRITER_SCOPE_ATTESTATION_MISSING", 503)
required_scope = {
"workspace_role": "task-scoped-isolated-writer",
"writes_enabled": True,
"allowed_branch": request["target_branch"],
"allowed_base_oid": request["base_oid"],
"allow_main_write": False,
"allow_tag_write": False,
"allow_baseline_write": False,
"allow_force_push": False,
}
for key, expected in required_scope.items():
if scope.get(key) != expected:
raise ProvisioningError("WRITER_SCOPE_ATTESTATION_MISMATCH", 503)
routes = scope.get("allowed_routes")
if routes != ["POST /api/commit-file"]:
raise ProvisioningError("WRITER_ROUTE_ATTESTATION_MISMATCH", 503)
negatives = result.get("negative_tests")
if not isinstance(negatives, dict) or any(
negatives.get(name) is not True for name in REQUIRED_NEGATIVE_TESTS
):
raise ProvisioningError("NEGATIVE_POLICY_TESTS_FAILED", 503)
return {
"branch": result["branch"],
"branch_oid": result["branch_oid"],
"base_oid": result["base_oid"],
"workspace_path": workspace,
"writer_scope": scope,
"negative_tests": {
name: True for name in sorted(REQUIRED_NEGATIVE_TESTS)
},
"grant_consumed": True,
"remote_ref_status": result.get("remote_ref_status"),
"remote_oid": result.get("remote_oid"),
"detached_writer": False,
"writer_config_path": None,
}
class AuthorizedTaskWriterProvisioner:
def __init__(
self,
config: ProvisioningConfig | None = None,
*,
clock=utc_now,
) -> None:
self.config = config or ProvisioningConfig.from_environment()
self.clock = clock
def provision(
self,
raw_request: Mapping[str, Any],
owner: OwnerContext,
) -> dict[str, Any]:
request = _validate_request(raw_request, owner=owner, config=self.config)
_prepare_state_root(self.config.state_root)
receipt_id = _safe_receipt_id(
request["task_id"], request["repository"], request["target_branch"]
)
receipt_path = self.config.state_root / "receipts" / f"{receipt_id}.json"
lock_path = self.config.state_root / "locks" / f"{receipt_id}.lock"
with open(lock_path, "a+b") as lock:
os.chmod(lock_path, 0o600)
fcntl.flock(lock.fileno(), fcntl.LOCK_EX)
return self._provision_locked(request, owner, receipt_path)
def _provision_locked(
self,
request: dict[str, Any],
owner: OwnerContext,
receipt_path: Path,
) -> dict[str, Any]:
receipt_exists = receipt_path.exists()
prestart = _prestart_git_check(
request, self.config, receipt_exists=receipt_exists
)
if receipt_exists:
receipt = _read_json(receipt_path)
expected_digest = sha256_json(
{
"task_id": request["task_id"],
"repository": request["repository"],
"base_ref": request["base_ref"],
"base_oid": request["base_oid"],
"target_branch": request["target_branch"],
"owner_authorization_reference": request[
"owner_authorization_reference"
],
}
)
if receipt.get("request_identity_sha256") != expected_digest:
raise ProvisioningError("IDEMPOTENCY_SCOPE_CONFLICT", 409)
return {
**receipt["public_result"],
"idempotent": True,
"prestart": prestart,
}
for path, name in (
(self.config.signer, "SIGNER"),
(self.config.validator, "VALIDATOR"),
(self.config.provisioner, "PROVISIONER"),
):
_verify_tool_contract(path, name, self.config)
now = self.clock()
plan, grant_request = _build_plan(request, owner, now=now)
temporary_root = Path(
tempfile.mkdtemp(
prefix="grant-",
dir=self.config.state_root,
)
)
os.chmod(temporary_root, 0o700)
try:
grant_dir = temporary_root / "grant"
grant_dir.mkdir(mode=0o700)
_atomic_write_json(grant_dir / "approval-plan.json", plan)
_atomic_write_json(grant_dir / "grant-request.json", grant_request)
signer_result_path = temporary_root / "signer-result.json"
validator_result_path = temporary_root / "validator-result.json"
provisioner_result_path = temporary_root / "provisioner-result.json"
try:
run_compatible_signer(
_tool_argv(self.config, self.config.signer),
grant_dir,
signer_result_path,
)
except Exception as exc:
raise ProvisioningError(
"SIGNER_COMPATIBILITY_FAILED",
503,
details={"error": type(exc).__name__},
) from exc
signer_result = _read_json(signer_result_path)
signer = _verify_signer_result(
signer_result,
approval_plan_id=plan["approval_plan_id"],
)
try:
validator_config_path = run_compatible_validator(
_tool_argv(self.config, self.config.validator),
grant_dir,
validator_result_path,
)
except Exception as exc:
raise ProvisioningError(
"VALIDATOR_COMPATIBILITY_FAILED",
503,
details={"error": type(exc).__name__},
) from exc
validator_result = _read_json(validator_result_path)
_verify_validator_result(
validator_result,
approval_plan_id=plan["approval_plan_id"],
grant_digest=signer["grant_digest_sha256"],
)
provisioner_env = dict(os.environ)
provisioner_env["AIOS_GRANT_ROOT_CONFIG"] = str(
validator_config_path
)
_run(
_tool_argv(
self.config,
self.config.provisioner,
"--grant-dir",
str(grant_dir),
"--result-json",
str(provisioner_result_path),
),
env=provisioner_env,
timeout=180,
)
provisioner_result = _read_json(provisioner_result_path)
provisioned = _verify_provisioner_result(
provisioner_result,
request,
grant_id=signer["grant_id"],
validator_config_path=validator_config_path,
)
post_branch_oid = _git(
self.config.authority_git_dir,
"show-ref",
"--verify",
"--hash",
f"refs/heads/{request['target_branch']}",
allowed_returncodes=(0, 1, 128),
)
if provisioned["detached_writer"]:
if post_branch_oid:
raise ProvisioningError("POST_PROVISION_TARGET_BRANCH_EXISTS", 503)
post_branch_oid = request["base_oid"]
elif post_branch_oid != request["base_oid"]:
raise ProvisioningError("POST_PROVISION_BRANCH_OID_MISMATCH", 503)
post_main_oid = _git(
self.config.authority_git_dir,
"rev-parse",
request["base_ref"],
)
if post_main_oid != request["base_oid"]:
raise ProvisioningError("POST_PROVISION_MAIN_DRIFT", 409)
public_result = {
"status": "READY_FOR_CODE_AUTHORING",
"task_id": request["task_id"],
"repository": request["repository"],
"approval_plan_id": plan["approval_plan_id"],
"grant_id": signer["grant_id"],
"grant_digest_sha256": signer["grant_digest_sha256"],
"signing_key_identity": signer["signing_key_identity"],
"branch": provisioned["branch"],
"branch_oid": provisioned["branch_oid"],
"base_oid": request["base_oid"],
"workspace_path": provisioned["workspace_path"],
"writer_scope": provisioned["writer_scope"],
"negative_tests": provisioned["negative_tests"],
"grant_consumed": True,
"main_unchanged": True,
"production_change_authorized": False,
"tag_change_authorized": False,
"baseline_change_authorized": False,
"idempotent": False,
}
request_identity = sha256_json(
{
"task_id": request["task_id"],
"repository": request["repository"],
"base_ref": request["base_ref"],
"base_oid": request["base_oid"],
"target_branch": request["target_branch"],
"owner_authorization_reference": request[
"owner_authorization_reference"
],
}
)
receipt = {
"schema": "aios.task-scoped-writer-provisioning-receipt.v1",
"request_identity_sha256": request_identity,
"created_at": iso_utc(self.clock()),
"public_result": _redacted(public_result),
"prestart": prestart,
"tool_evidence": {
"signer": {
"status": signer_result.get("status"),
"grant_digest_sha256": signer["grant_digest_sha256"],
"signing_key_identity": signer["signing_key_identity"],
},
"validator": {
"status": validator_result.get("status"),
"checks": validator_result.get("checks"),
},
"provisioner": {
"status": provisioner_result.get("status"),
"grant_consumed": provisioner_result.get("grant_consumed"),
},
},
}
_atomic_write_json(receipt_path, receipt)
return {**public_result, "prestart": prestart}
finally:
shutil.rmtree(temporary_root, ignore_errors=True)
@@ -0,0 +1,89 @@
"""Durable Owner/GPT notification outbox with retry and acknowledgement."""
from __future__ import annotations
import json
import sqlite3
import time
import uuid
from dataclasses import dataclass
from pathlib import Path
from typing import Any, Callable, Mapping
@dataclass(frozen=True)
class OutboxEvent:
event_id: str
recipient: str
kind: str
payload: Mapping[str, Any]
attempts: int
available_at: float
acknowledged: bool
class DurableOutbox:
RECIPIENTS = ("owner", "gpt")
def __init__(self, path: str | Path) -> None:
self.path = str(path)
self.db = sqlite3.connect(self.path)
self.db.execute("PRAGMA journal_mode=WAL")
self.db.execute("CREATE TABLE IF NOT EXISTS outbox (event_id TEXT PRIMARY KEY, recipient TEXT NOT NULL, kind TEXT NOT NULL, payload TEXT NOT NULL, attempts INTEGER NOT NULL DEFAULT 0, available_at REAL NOT NULL, delivered_at REAL, acknowledged INTEGER NOT NULL DEFAULT 0, last_error TEXT)")
self.db.commit()
def enqueue(self, kind: str, payload: Mapping[str, Any], *, event_id: str | None = None, recipients: tuple[str, ...] = RECIPIENTS) -> list[str]:
ids: list[str] = []
for recipient in recipients:
if recipient not in self.RECIPIENTS:
raise ValueError("recipient is outside the governed notification plane")
eid = event_id or str(uuid.uuid4())
if event_id is not None and len(recipients) > 1:
eid = f"{event_id}:{recipient}"
self.db.execute("INSERT OR IGNORE INTO outbox(event_id,recipient,kind,payload,available_at) VALUES(?,?,?,?,?)", (eid, recipient, kind, json.dumps(dict(payload), sort_keys=True), time.time()))
ids.append(eid)
self.db.commit()
return ids
def due(self, *, now: float | None = None, limit: int = 20) -> list[OutboxEvent]:
now = time.time() if now is None else now
rows = self.db.execute("SELECT event_id,recipient,kind,payload,attempts,available_at,acknowledged FROM outbox WHERE delivered_at IS NULL AND available_at<=? ORDER BY available_at,event_id LIMIT ?", (now, limit)).fetchall()
return [OutboxEvent(r[0], r[1], r[2], json.loads(r[3]), r[4], r[5], bool(r[6])) for r in rows]
def delivered(self, event_id: str, *, now: float | None = None) -> None:
self.db.execute("UPDATE outbox SET delivered_at=? WHERE event_id=?", (time.time() if now is None else now, event_id))
self.db.commit()
def failed(self, event_id: str, error: str, *, now: float | None = None, max_backoff: float = 300.0) -> None:
now = time.time() if now is None else now
row = self.db.execute("SELECT attempts FROM outbox WHERE event_id=?", (event_id,)).fetchone()
attempts = int(row[0]) + 1 if row else 1
delay = min(max_backoff, 2 ** min(attempts, 8))
self.db.execute("UPDATE outbox SET attempts=?,available_at=?,last_error=? WHERE event_id=?", (attempts, now + delay, error[:200], event_id))
self.db.commit()
def acknowledge(self, event_id: str) -> None:
self.db.execute("UPDATE outbox SET acknowledged=1 WHERE event_id=?", (event_id,))
self.db.commit()
def close(self) -> None:
self.db.close()
class NotificationPlane:
def __init__(self, outbox: DurableOutbox) -> None:
self.outbox = outbox
def emit(self, kind: str, payload: Mapping[str, Any], *, event_id: str | None = None) -> list[str]:
return self.outbox.enqueue(kind, payload, event_id=event_id)
def deliver_once(self, send: Callable[[OutboxEvent], None], *, now: float | None = None) -> list[str]:
delivered: list[str] = []
for event in self.outbox.due(now=now):
try:
send(event)
except Exception as exc:
self.outbox.failed(event.event_id, type(exc).__name__, now=now)
else:
self.outbox.delivered(event.event_id, now=now)
delivered.append(event.event_id)
return delivered
+491
View File
@@ -0,0 +1,491 @@
"""Fail-closed, branch-bound repair leases for governed AIOS GitOps work."""
from __future__ import annotations
import hashlib
import json
import os
import posixpath
import re
import secrets
import subprocess
import tempfile
import uuid
from dataclasses import asdict, dataclass, field
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Iterable, Mapping, Protocol, Sequence
TASK_ID = "V2R1C6_GOVERNED_LOCAL_BOOTSTRAP"
MAX_LEASE_SECONDS = 86_400
OID_RE = re.compile(r"^[0-9a-fA-F]{40}$")
DENIED_BRANCHES = {"main", "master", "HEAD"}
PROTECTED_PREFIXES = (
".git",
"VERSION",
"versions/V2R1C5.json",
"artifacts/V2R1C5",
"baseline",
"config/gitops.tokens",
"gitops.tokens",
"secrets",
"credentials",
)
SECRET_PATTERNS = (
re.compile(r"(?i)-----BEGIN [A-Z ]*PRIVATE KEY-----"),
re.compile(r"(?i)(?:api[_-]?key|secret|password|token)\s*[:=]\s*\S+"),
re.compile(r"\b(?:sk|HPUA|SDC)-[A-Za-z0-9_-]{16,}\b"),
)
class LeaseError(Exception):
"""Raised for any invalid, stale, expired, or prohibited lease action."""
class GitProbeProtocol(Protocol):
"""Read-only Git seam used by the lease manager."""
def branch_head(self, branch: str) -> str: ...
def parent(self, commit: str) -> str: ...
def is_ancestor(self, older: str, newer: str) -> bool: ...
def changed_paths(self, commit: str) -> tuple[str, ...]: ...
def commits_between(self, older: str, newer: str) -> tuple[str, ...]: ...
def _now() -> datetime:
return datetime.now(timezone.utc)
def _ts(value: datetime) -> str:
return value.astimezone(timezone.utc).replace(microsecond=0).isoformat().replace("+00:00", "Z")
def _parse_ts(value: str) -> datetime:
if value.endswith("Z"):
value = value[:-1] + "+00:00"
parsed = datetime.fromisoformat(value)
return parsed if parsed.tzinfo else parsed.replace(tzinfo=timezone.utc)
def _oid(value: str, field: str = "oid") -> str:
if not isinstance(value, str) or not OID_RE.fullmatch(value):
raise LeaseError(f"{field} must be a full commit OID")
return value.lower()
def normalize_path(path: str) -> str:
"""Normalize a repository-relative path without resolving through the FS."""
if not isinstance(path, str) or not path.strip():
raise LeaseError("path must be a non-empty string")
value = path.replace("\\", "/").strip()
if value.startswith("/") or "\x00" in value:
raise LeaseError("path must be relative")
parts = value.split("/")
if any(part in {"", ".", ".."} for part in parts):
raise LeaseError("path contains a prohibited component")
normalized = posixpath.normpath(value)
if normalized == "." or normalized.startswith("../"):
raise LeaseError("path is not normalized")
return normalized
def path_is_protected(path: str) -> bool:
normalized = normalize_path(path)
lower = normalized.lower()
return any(
lower == prefix.lower() or lower.startswith(prefix.lower().rstrip("/") + "/")
for prefix in PROTECTED_PREFIXES
) or lower.startswith("refs/")
def contains_secret_like_content(content: str) -> bool:
return any(pattern.search(content or "") for pattern in SECRET_PATTERNS)
def validate_paths(paths: Iterable[str], allowed_paths: Sequence[str]) -> tuple[str, ...]:
allowed = tuple(normalize_path(path) for path in allowed_paths)
if not allowed:
raise LeaseError("allowed_paths must not be empty")
result: list[str] = []
for raw in paths:
path = normalize_path(raw)
if path_is_protected(path):
raise LeaseError("protected path is not lease-writable")
if not any(path == prefix or path.startswith(prefix + "/") for prefix in allowed):
raise LeaseError("path is outside the lease allowlist")
if path not in result:
result.append(path)
if not result:
raise LeaseError("at least one changed path is required")
return tuple(result)
@dataclass
class CommitEntry:
commit_oid: str
parent_oid: str
paths: tuple[str, ...]
recorded_at: str
@dataclass
class RepairLease:
lease_id: str
task_id: str
owner_identity: str
branch: str
description: str
allowed_paths: tuple[str, ...]
initial_head_oid: str
current_head_oid: str
created_at: str
expires_at: str
max_expires_at: str
state: str = "ACTIVE"
revision: int = 0
commits: list[CommitEntry] = field(default_factory=list)
attached: bool = False
def to_json(self) -> dict[str, Any]:
payload = asdict(self)
payload["allowed_paths"] = list(self.allowed_paths)
payload["commits"] = [asdict(entry) for entry in self.commits]
return payload
@classmethod
def from_json(cls, raw: dict[str, Any]) -> "RepairLease":
commits = tuple(
CommitEntry(
commit_oid=_oid(item["commit_oid"], "commit_oid"),
parent_oid=_oid(item["parent_oid"], "parent_oid"),
paths=tuple(normalize_path(path) for path in item["paths"]),
recorded_at=str(item["recorded_at"]),
)
for item in raw.get("commits", [])
)
return cls(
lease_id=str(raw["lease_id"]),
task_id=str(raw["task_id"]),
owner_identity=str(raw["owner_identity"]),
branch=str(raw["branch"]),
description=str(raw["description"]),
allowed_paths=tuple(normalize_path(path) for path in raw["allowed_paths"]),
initial_head_oid=_oid(raw["initial_head_oid"], "initial_head_oid"),
current_head_oid=_oid(raw["current_head_oid"], "current_head_oid"),
created_at=str(raw["created_at"]),
expires_at=str(raw["expires_at"]),
max_expires_at=str(raw["max_expires_at"]),
state=str(raw.get("state", "ACTIVE")),
revision=int(raw.get("revision", 0)),
commits=list(commits),
attached=bool(raw.get("attached", False)),
)
class LeaseStore:
"""Atomic JSON store for restart-recoverable lease state."""
def __init__(self, directory: str | os.PathLike[str]) -> None:
self.directory = Path(directory)
self.directory.mkdir(parents=True, exist_ok=True)
if self.directory.is_symlink():
raise LeaseError("lease store must not be a symlink")
os.chmod(self.directory, 0o700)
def _path(self, lease_id: str) -> Path:
if not re.fullmatch(r"lease-[0-9a-f]{32}", lease_id):
raise LeaseError("invalid lease id")
return self.directory / f"{lease_id}.json"
def create(self, lease: RepairLease) -> None:
path = self._path(lease.lease_id)
if path.exists() or path.is_symlink():
raise LeaseError("lease already exists")
self._write(path, lease.to_json(), exclusive=True)
def load(self, lease_id: str) -> RepairLease:
path = self._path(lease_id)
try:
raw = json.loads(path.read_text(encoding="utf-8"))
except (OSError, json.JSONDecodeError, TypeError) as exc:
raise LeaseError("lease state is unavailable or invalid") from exc
if not isinstance(raw, dict):
raise LeaseError("lease state must be an object")
return RepairLease.from_json(raw)
def save(self, lease: RepairLease) -> None:
self._write(self._path(lease.lease_id), lease.to_json())
@staticmethod
def _write(path: Path, payload: dict[str, Any], *, exclusive: bool = False) -> None:
flags = os.O_WRONLY | os.O_CREAT | (os.O_EXCL if exclusive else 0)
fd, temp_name = tempfile.mkstemp(prefix=".lease-", dir=path.parent)
try:
os.fchmod(fd, 0o600)
with os.fdopen(fd, "w", encoding="utf-8") as stream:
json.dump(payload, stream, ensure_ascii=True, sort_keys=True, separators=(",", ":"))
stream.write("\n")
if exclusive:
os.link(temp_name, path)
os.unlink(temp_name)
else:
os.replace(temp_name, path)
os.chmod(path, 0o600)
except Exception:
try:
os.unlink(temp_name)
except FileNotFoundError:
pass
raise
class GitProbe:
"""Read-only Git implementation; it never changes a ref or working tree."""
def __init__(self, bare_repo: str | os.PathLike[str]) -> None:
self.bare_repo = Path(bare_repo)
def _run(self, args: list[str]) -> str:
result = subprocess.run(
["git", "--git-dir", str(self.bare_repo), *args],
text=True,
capture_output=True,
check=False,
)
if result.returncode != 0:
raise LeaseError("read-only Git probe failed")
return result.stdout.strip()
def branch_head(self, branch: str) -> str:
return _oid(self._run(["rev-parse", f"refs/heads/{branch}"]), "branch_head")
def parent(self, commit: str) -> str:
return _oid(self._run(["rev-parse", f"{_oid(commit)}^1"]), "parent")
def is_ancestor(self, older: str, newer: str) -> bool:
old, new = _oid(older), _oid(newer)
result = subprocess.run(
["git", "--git-dir", str(self.bare_repo), "merge-base", "--is-ancestor", old, new],
check=False,
)
return result.returncode == 0
def changed_paths(self, commit: str) -> tuple[str, ...]:
raw = self._run(["diff-tree", "--no-commit-id", "--name-only", "-r", _oid(commit)])
return tuple(normalize_path(line) for line in raw.splitlines() if line.strip())
def commits_between(self, older: str, newer: str) -> tuple[str, ...]:
raw = self._run(["rev-list", "--reverse", f"{_oid(older)}..{_oid(newer)}"])
return tuple(_oid(line, "commit") for line in raw.splitlines() if line.strip())
class RepairLeaseManager:
"""Issue, recover, renew, and close branch-bound repair leases."""
def __init__(self, store: LeaseStore, git: GitProbeProtocol, *, clock=_now) -> None:
self.store = store
self.git = git
self.clock = clock
def _require_owner(self, lease: RepairLease, owner_identity: str) -> None:
if owner_identity != lease.owner_identity:
raise LeaseError("lease identity mismatch")
def _active(self, lease: RepairLease) -> RepairLease:
now = self.clock()
if lease.state != "ACTIVE":
raise LeaseError(f"lease is {lease.state.lower()}")
if now >= _parse_ts(lease.expires_at):
lease.state = "EXPIRED"
self.store.save(lease)
raise LeaseError("lease expired")
return lease
def issue(
self,
*,
task_id: str,
owner_identity: str,
branch: str,
description: str,
allowed_paths: Sequence[str],
ttl_seconds: int = 3600,
) -> RepairLease:
if task_id != TASK_ID:
raise LeaseError("unexpected task id")
if not owner_identity or not description or len(description) > 240:
raise LeaseError("owner identity and description are required")
if "\n" in description or "\r" in description:
raise LeaseError("description must be a single line")
if branch in DENIED_BRANCHES or branch.startswith(("main/", "refs/", "tags/")):
raise LeaseError("protected branch cannot receive a repair lease")
if not isinstance(ttl_seconds, int) or not 0 < ttl_seconds <= MAX_LEASE_SECONDS:
raise LeaseError("lease TTL exceeds the maximum")
allowed = tuple(normalize_path(path) for path in allowed_paths)
if any(path_is_protected(path) for path in allowed):
raise LeaseError("protected path in lease allowlist")
head = self.git.branch_head(branch)
created = self.clock()
expires = created + timedelta(seconds=ttl_seconds)
max_expires = created + timedelta(seconds=MAX_LEASE_SECONDS)
lease_id = "lease-" + secrets.token_hex(16)
lease = RepairLease(
lease_id=lease_id,
task_id=task_id,
owner_identity=owner_identity,
branch=branch,
description=description,
allowed_paths=allowed,
initial_head_oid=head,
current_head_oid=head,
created_at=_ts(created),
expires_at=_ts(expires),
max_expires_at=_ts(max_expires),
)
self.store.create(lease)
return lease
def attach_existing_branch(
self, lease_id: str, *, owner_identity: str, expected_head_oid: str
) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
current = self.git.branch_head(lease.branch)
expected = _oid(expected_head_oid, "expected_head_oid")
if current != expected or current != lease.current_head_oid:
raise LeaseError("branch head does not match the lease")
lease.attached = True
lease.revision += 1
self.store.save(lease)
return lease
def authorize_commit(
self,
lease_id: str,
*,
owner_identity: str,
expected_head_oid: str,
paths: Sequence[str],
content: Mapping[str, str] | None = None,
) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
expected = _oid(expected_head_oid, "expected_head_oid")
current = self.git.branch_head(lease.branch)
if current != expected or current != lease.current_head_oid:
raise LeaseError("stale or externally advanced branch head")
if not lease.attached:
raise LeaseError("lease is not attached")
validate_paths(paths, lease.allowed_paths)
if content and any(contains_secret_like_content(value) for value in content.values()):
raise LeaseError("secret-like content rejected")
return lease
def record_commit(
self,
lease_id: str,
*,
owner_identity: str,
expected_head_oid: str,
new_head_oid: str,
paths: Sequence[str],
) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
new_head = _oid(new_head_oid, "new_head_oid")
expected = _oid(expected_head_oid, "expected_head_oid")
if not lease.attached or lease.current_head_oid != expected:
raise LeaseError("stale lease head")
if self.git.branch_head(lease.branch) != new_head:
raise LeaseError("branch head does not match committed head")
validate_paths(paths, lease.allowed_paths)
if self.git.parent(new_head) != expected or not self.git.is_ancestor(expected, new_head):
raise LeaseError("commit is not a fast-forward child of the expected head")
actual_paths = self.git.changed_paths(new_head)
allowed = validate_paths(actual_paths, lease.allowed_paths)
requested = set(validate_paths(paths, lease.allowed_paths))
if not set(allowed).issubset(requested):
raise LeaseError("commit changed an unrequested path")
lease.current_head_oid = new_head
lease.revision += 1
lease.commits.append(
CommitEntry(
commit_oid=new_head,
parent_oid=expected,
paths=allowed,
recorded_at=_ts(self.clock()),
)
)
self.store.save(lease)
return lease
def recover(self, lease_id: str, *, owner_identity: str) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
current = self.git.branch_head(lease.branch)
if current == lease.current_head_oid:
return lease
if not self.git.is_ancestor(lease.current_head_oid, current):
raise LeaseError("branch moved non-fast-forward")
commits = self.git.commits_between(lease.current_head_oid, current)
for commit in commits:
actual = self.git.changed_paths(commit)
validate_paths(actual, lease.allowed_paths)
lease.current_head_oid = current
lease.revision += 1
self.store.save(lease)
return lease
def renew(self, lease_id: str, *, owner_identity: str, ttl_seconds: int) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
if not isinstance(ttl_seconds, int) or ttl_seconds <= 0:
raise LeaseError("renewal TTL must be positive")
candidate = self.clock() + timedelta(seconds=ttl_seconds)
maximum = _parse_ts(lease.max_expires_at)
if candidate > maximum:
raise LeaseError("renewal exceeds maximum lease lifetime")
lease.expires_at = _ts(candidate)
lease.revision += 1
self.store.save(lease)
return lease
def complete(self, lease_id: str, *, owner_identity: str) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
lease.state = "COMPLETED"
lease.revision += 1
self.store.save(lease)
return lease
def revoke(self, lease_id: str, *, owner_identity: str) -> RepairLease:
lease = self._active(self.store.load(lease_id))
self._require_owner(lease, owner_identity)
lease.state = "REVOKED"
lease.revision += 1
self.store.save(lease)
return lease
def sanitized_lease_view(lease: RepairLease) -> dict[str, Any]:
"""Return lease metadata only; no content, token, or credential values."""
return {
"lease_id": lease.lease_id,
"task_id": lease.task_id,
"owner_identity": lease.owner_identity,
"branch": lease.branch,
"description": lease.description,
"initial_head_oid": lease.initial_head_oid,
"current_head_oid": lease.current_head_oid,
"created_at": lease.created_at,
"expires_at": lease.expires_at,
"state": lease.state,
"revision": lease.revision,
"commit_count": len(lease.commits),
"attached": lease.attached,
}
+67
View File
@@ -0,0 +1,67 @@
"""Runtime deployment-state provider for governed health contracts."""
from __future__ import annotations
import datetime as dt
import json
import os
import re
from dataclasses import dataclass
from pathlib import Path
from typing import Any, Mapping
_OID = re.compile(r"^[0-9a-f]{40}$")
_SHA = re.compile(r"^[0-9a-f]{64}$")
_STATES = {"STAGED", "ACTIVE", "ROLLED_BACK"}
@dataclass(frozen=True)
class RuntimeDeploymentState:
version: str
commit_oid: str
artifact_sha256: str
deployment_state: str
@property
def production_deployed(self) -> bool:
return self.deployment_state == "ACTIVE"
def health_payload(self) -> dict[str, Any]:
return {
"status": "ok",
"version": self.version,
"commit_oid": self.commit_oid,
"artifact_sha256": self.artifact_sha256,
"deployment_state": self.deployment_state,
"production_deployed": self.production_deployed,
"health_timestamp": dt.datetime.now(dt.timezone.utc).isoformat(),
}
@classmethod
def from_environment(cls, environ: Mapping[str, str] | None = None) -> "RuntimeDeploymentState":
env = dict(environ or os.environ)
values: dict[str, Any] = {}
state_file = env.get("DEPLOYMENT_STATE_FILE", "").strip()
if state_file:
path = Path(state_file)
if path.is_file() and not path.is_symlink():
loaded = json.loads(path.read_text(encoding="utf-8"))
if not isinstance(loaded, dict):
raise ValueError("deployment_state_file_object_required")
values.update(loaded)
file_deployment_state = values.get("deployment_state")
values.update({k: v for k, v in {
"version": env.get("RELEASE_VERSION", values.get("version", "")),
"commit_oid": env.get("TARGET_OID", values.get("commit_oid", "")),
"artifact_sha256": env.get("ARTIFACT_SHA256", values.get("artifact_sha256", "")),
"deployment_state": env.get("DEPLOYMENT_STATE", values.get("deployment_state", "")),
}.items() if v != ""})
if file_deployment_state not in (None, ""):
values["deployment_state"] = file_deployment_state
version = str(values.get("version", ""))
commit_oid = str(values.get("commit_oid", ""))
artifact_sha256 = str(values.get("artifact_sha256", ""))
deployment_state = str(values.get("deployment_state", ""))
if not version or not _OID.fullmatch(commit_oid) or not _SHA.fullmatch(artifact_sha256):
raise ValueError("deployment_binding_invalid")
if deployment_state not in _STATES:
raise ValueError("deployment_state_invalid")
return cls(version, commit_oid, artifact_sha256, deployment_state)
+473
View File
@@ -0,0 +1,473 @@
"""Shared primitives for the V2R1C6 governed local writer contract."""
from __future__ import annotations
import hashlib
import json
import os
import posixpath
import re
import socket
import stat
import struct
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Mapping, Sequence
TASK_ID = "V2R1C6_GOVERNED_LOCAL_BOOTSTRAP"
REPO = "aios/aios"
PRINCIPAL = "governed-local-bootstrap"
OIDSIZ = 40
AUTHORIZATION_TTL = timedelta(hours=2)
class GovernedLocalWriterError(Exception):
"""Base error for governed-local-writer validation and runtime failures."""
class AuthorizationError(GovernedLocalWriterError):
"""Raised when authorization input is invalid, expired, or tampered."""
@dataclass(frozen=True)
class BranchCreatePolicy:
"""Exact branch creation contract."""
branch: str
base_oid: str
head_oid: str
@dataclass(frozen=True)
class BranchAppendPolicy:
"""Exact append-commit contract for candidate branch updates."""
branch: str
expected_head_oid: str | None
allowed_paths: tuple[str, ...]
@dataclass(frozen=True)
class BranchPolicy:
"""Per-branch operation policies loaded from authorization."""
create_branch: BranchCreatePolicy | None
append_commit: BranchAppendPolicy | None
@dataclass(frozen=True)
class GovernedAuthorization:
"""Parsed and validated authorization object."""
task_id: str
repo: str
created_at: datetime
expires_at: datetime
authorization_sha256: str
one_time_use: bool
branch_operations: dict[str, BranchPolicy]
denied_operations: tuple[str, ...]
@property
def create_branches(self) -> set[str]:
"""Branches that are allowed for branch-creation operations."""
return {
branch
for branch, policy in self.branch_operations.items()
if policy.create_branch is not None
}
@property
def append_branches(self) -> set[str]:
"""Branches that are allowed for commit-append operations."""
return {
branch
for branch, policy in self.branch_operations.items()
if policy.append_commit is not None
}
@dataclass(frozen=True)
class PeerIdentity:
"""Verified peer identity from SO_PEERCRED."""
pid: int
uid: int
gid: int
executable: str | None
@dataclass(frozen=True)
class FileChange:
"""Single file modification payload."""
path: str
content: str
@dataclass(frozen=True)
class CreateBranchRequest:
"""Validated create-branch request payload."""
branch: str
base_oid: str
head_oid: str | None
authorization_digest: str
@dataclass(frozen=True)
class AppendCommitRequest:
"""Validated append-commit request payload."""
branch: str
expected_head_oid: str
changes: tuple[FileChange, ...]
authorization_digest: str
commit_message: str = "chore(governance): append governed commit"
_OID_RE = re.compile(r"^[0-9a-fA-F]{40}$")
_DENIED_PATH_SEGMENTS = (
"main",
"tag",
"baseline",
"credential",
"credentials",
"secret",
"registry",
)
_SECRET_PATTERNS = (
re.compile(r"(?i)-----BEGIN [A-Z ]*PRIVATE KEY-----"),
re.compile(r"(?i)api[_-]?key\s*[:=]\s*\S+"),
re.compile(r"(?i)secret\s*[:=]\s*\S+"),
re.compile(r"(?i)password\s*[:=]\s*\S+"),
re.compile(r"(?i)token\s*[:=]\s*\S+"),
)
_SAFE_SECRET_MARKERS = re.compile(
r"(?i)(?:api[_-]?key|secret|password|token)\s*[:=]\s*(?:REDACTED|TEST_FIXTURE|NOT_A_SECRET)"
)
def canonicalize_authorization_payload(payload: Mapping[str, Any]) -> str:
"""Serialize authorization payload without the signature hash, in canonical form."""
canonical = dict(payload)
canonical.pop("authorization_sha256", None)
return json.dumps(
canonical, sort_keys=True, separators=(",", ":"), ensure_ascii=True
)
def compute_authorization_sha256(payload: Mapping[str, Any]) -> str:
"""Compute canonical sha256 over authorization payload."""
digest = hashlib.sha256()
digest.update(canonicalize_authorization_payload(payload).encode("utf-8"))
return digest.hexdigest()
def _assert_authorization_file_permissions(path: Path) -> None:
"""Fail-closed checks for root-only secret-like authorization files."""
if path.is_symlink():
raise AuthorizationError("authorization path must not be a symlink")
st = path.stat()
if not stat.S_ISREG(st.st_mode):
raise AuthorizationError("authorization path must be a regular file")
if (st.st_mode & 0o777) != 0o600:
raise AuthorizationError("authorization file mode must be exactly 0600")
if hasattr(st, "st_uid") and os.name != "nt" and st.st_uid != 0:
raise AuthorizationError("authorization file owner must be uid 0")
def _parse_iso8601(value: str) -> datetime:
"""Parse ISO 8601 timestamps used by authorization payloads."""
try:
if value.endswith("Z"):
value = value[:-1] + "+00:00"
parsed = datetime.fromisoformat(value)
except (TypeError, ValueError) as exc:
raise AuthorizationError("created_at must be an ISO 8601 timestamp") from exc
if parsed.tzinfo is None:
parsed = parsed.replace(tzinfo=timezone.utc)
return parsed
def _require_type(value: Any, expected: type, *, key: str) -> None:
if not isinstance(value, expected):
raise AuthorizationError(f"{key} must be {expected.__name__}")
def _parse_oid(value: Any, *, field: str) -> str:
if not isinstance(value, str):
raise AuthorizationError(f"{field} must be a string")
if not _OID_RE.fullmatch(value):
raise AuthorizationError(f"{field} must be a 40-char hex object id")
return value.lower()
def _parse_allowed_paths(paths: Any) -> tuple[str, ...]:
if not isinstance(paths, list) or not paths:
raise AuthorizationError("allowed_paths must be a non-empty list")
normalized: list[str] = []
for path in paths:
if not isinstance(path, str) or not path.strip():
raise AuthorizationError("allowed_paths entries must be non-empty strings")
normalized.append(path.strip())
return tuple(normalized)
def parse_authorization(
raw: Mapping[str, Any], *, now: datetime | None = None
) -> GovernedAuthorization:
"""Validate an authorization payload and return a typed object."""
_require_type(raw.get("task_id"), str, key="task_id")
if raw["task_id"] != TASK_ID:
raise AuthorizationError("unexpected task_id")
_require_type(raw.get("repo"), str, key="repo")
if raw["repo"] != REPO:
raise AuthorizationError("unexpected repo value")
created_at = _parse_iso8601(raw.get("created_at"))
check_time = now or datetime.now(tz=timezone.utc)
if created_at.tzinfo is None:
created_at = created_at.replace(tzinfo=timezone.utc)
if created_at - check_time > timedelta(seconds=30):
raise AuthorizationError("authorization created_at is in the future")
if check_time - created_at > AUTHORIZATION_TTL:
raise AuthorizationError("authorization expired")
raw_expires_at = raw.get("expires_at")
expires_at = (
_parse_iso8601(raw_expires_at)
if raw_expires_at is not None
else created_at + AUTHORIZATION_TTL
)
if expires_at <= check_time:
raise AuthorizationError("authorization expiry is in the past")
if expires_at > created_at + AUTHORIZATION_TTL:
raise AuthorizationError("authorization expiry exceeds two-hour maximum")
_require_type(raw.get("one_time_use"), bool, key="one_time_use")
if raw["one_time_use"] is not True:
raise AuthorizationError("one_time_use must be true")
_require_type(raw.get("authorization_sha256"), str, key="authorization_sha256")
expected_hash = compute_authorization_sha256(raw)
provided_hash = str(raw["authorization_sha256"]).lower()
if provided_hash != expected_hash:
raise AuthorizationError("authorization_sha256 mismatch")
op_map = raw.get("allowed_branch_operations")
if not isinstance(op_map, dict) or not op_map:
raise AuthorizationError("allowed_branch_operations must be a non-empty object")
branch_operations: dict[str, BranchPolicy] = {}
for branch, operation in op_map.items():
if not isinstance(branch, str) or not branch.strip():
raise AuthorizationError("branch names must be non-empty strings")
if not isinstance(operation, dict):
raise AuthorizationError("operation policy must be an object")
create_block = operation.get("create_branch")
append_block = operation.get("append_commit")
create_policy: BranchCreatePolicy | None = None
append_policy: BranchAppendPolicy | None = None
if create_block is not None:
if not isinstance(create_block, dict):
raise AuthorizationError("create_branch block must be an object")
raw_head_oid = create_block.get("head_oid")
create_policy = BranchCreatePolicy(
branch=branch,
base_oid=_parse_oid(create_block.get("base_oid"), field="create_branch.base_oid"),
head_oid=(
_parse_oid(raw_head_oid, field="create_branch.head_oid")
if raw_head_oid is not None
else None
),
)
if append_block is not None:
if not isinstance(append_block, dict):
raise AuthorizationError("append_commit block must be an object")
raw_expected_head_oid = append_block.get("expected_head_oid")
append_policy = BranchAppendPolicy(
branch=branch,
expected_head_oid=(
_parse_oid(
raw_expected_head_oid,
field="append_commit.expected_head_oid",
)
if raw_expected_head_oid is not None
else None
),
allowed_paths=_parse_allowed_paths(append_block.get("allowed_paths")),
)
if create_policy is None and append_policy is None:
raise AuthorizationError("branch policy must define at least one operation")
branch_operations[branch] = BranchPolicy(
create_branch=create_policy, append_commit=append_policy
)
denied_raw = raw.get("denied_operations", ())
if not isinstance(denied_raw, list):
raise AuthorizationError("denied_operations must be a list")
denied_operations: list[str] = []
for value in denied_raw:
if not isinstance(value, str):
raise AuthorizationError("denied operations must be strings")
denied_operations.append(value.lower())
return GovernedAuthorization(
task_id=raw["task_id"],
repo=raw["repo"],
created_at=created_at,
expires_at=expires_at,
authorization_sha256=provided_hash,
one_time_use=True,
branch_operations=branch_operations,
denied_operations=tuple(denied_operations),
)
def load_authorization_file(
path: str | os.PathLike[str], *, now: datetime | None = None
) -> GovernedAuthorization:
"""Load and validate the root-only authorization file."""
auth_path = Path(path)
_assert_authorization_file_permissions(auth_path)
try:
raw_text = auth_path.read_text(encoding="utf-8")
except (OSError, ValueError) as exc:
raise AuthorizationError("unable to read authorization file") from exc
try:
raw_obj = json.loads(raw_text)
except json.JSONDecodeError as exc:
raise AuthorizationError("authorization must be JSON") from exc
if not isinstance(raw_obj, dict):
raise AuthorizationError("authorization top-level must be an object")
return parse_authorization(raw_obj, now=now)
def _resolve_peer_executable(pid: int) -> str | None:
"""Resolve peer executable from /proc/<pid>/exe when available."""
if pid <= 0:
return None
exe_path = f"/proc/{pid}/exe"
try:
return os.readlink(exe_path)
except OSError:
return None
def extract_peer_identity(
connection: socket.socket, *, require_root_uid: bool = True
) -> PeerIdentity:
"""Validate SO_PEERCRED and return peer identity."""
if not hasattr(socket, "SO_PEERCRED"):
raise AuthorizationError("SO_PEERCRED not available on this platform")
data = connection.getsockopt(
socket.SOL_SOCKET,
socket.SO_PEERCRED,
struct.calcsize("3i"),
)
if len(data) != struct.calcsize("3i"):
raise AuthorizationError("SO_PEERCRED payload size mismatch")
pid, uid, gid = struct.unpack("3i", data)
if require_root_uid and uid != 0:
raise PermissionError("peer uid must be 0")
return PeerIdentity(
pid=pid,
uid=uid,
gid=gid,
executable=_resolve_peer_executable(pid),
)
def _normalize_file_path(path: str) -> str:
normalized = posixpath.normpath(path.replace("\\", "/").strip())
return normalized.lstrip("/")
def _path_has_disallowed_segment(path: str) -> bool:
lowered = path.lower()
if ".." in path.split("/"):
return True
parts = [p for p in lowered.split("/") if p]
for part in parts:
for segment in _DENIED_PATH_SEGMENTS:
if part == segment or part.startswith(f"{segment}.") or segment in part:
return True
return False
def _is_allowed_path(path: str, allowed_paths: Sequence[str]) -> bool:
normalized = _normalize_file_path(path)
for prefix in allowed_paths:
normalized_prefix = _normalize_file_path(prefix)
if normalized == normalized_prefix:
return True
if normalized.startswith(normalized_prefix + "/"):
return True
return False
def contains_secret_like_content(content: str) -> bool:
"""Heuristic detector for obvious secret-like strings."""
content = _SAFE_SECRET_MARKERS.sub("", content or "")
for pattern in _SECRET_PATTERNS:
if pattern.search(content):
return True
return False
def validate_file_changes(
changes: Any, allowed_paths: Sequence[str]
) -> tuple[FileChange, ...]:
"""Validate append-commit file changes against path and content policy."""
if not isinstance(changes, list):
raise AuthorizationError("changes must be a list")
result: list[FileChange] = []
for change in changes:
if not isinstance(change, Mapping):
raise AuthorizationError("each file change must be an object")
raw_path = change.get("path")
raw_content = change.get("content")
if not isinstance(raw_path, str) or not raw_path.strip():
raise AuthorizationError("file change path must be a non-empty string")
if raw_path.startswith(("/", "\\")):
raise AuthorizationError("file change path must be relative")
if not isinstance(raw_content, str):
raise AuthorizationError("file change content must be a string")
normalized = _normalize_file_path(raw_path)
if not normalized or normalized.startswith("../") or normalized == "..":
raise AuthorizationError("file paths must be normalized relative paths")
if not _is_allowed_path(normalized, allowed_paths):
raise AuthorizationError("file path outside allowed_paths")
if _path_has_disallowed_segment(normalized):
raise AuthorizationError("file path uses denied segment")
if contains_secret_like_content(raw_content):
raise AuthorizationError("file content failed secret-like scan")
result.append(FileChange(path=normalized, content=raw_content))
return tuple(result)
+47
View File
@@ -0,0 +1,47 @@
"""Supplement Gitea native protection when review identity is not expressible."""
from __future__ import annotations
from dataclasses import dataclass
from typing import Any, Mapping, Sequence
class MergeGateDenied(RuntimeError):
pass
@dataclass(frozen=True)
class MergeGateDecision:
allowed: bool
owner_review_id: int
required_contexts: tuple[str, ...]
def validate_protected_merge(
pull_request: Mapping[str, Any],
reviews: Sequence[Mapping[str, Any]],
statuses: Mapping[str, str],
*,
required_contexts: Sequence[str],
unresolved_conversations: int,
owner_login: str = "owner",
owner_id: int = 2,
) -> MergeGateDecision:
if pull_request.get("state") != "open" or pull_request.get("mergeable") is not True:
raise MergeGateDenied("pull_request_not_mergeable")
author = (pull_request.get("user") or {}).get("login")
owner_review = None
for review in reviews:
user = review.get("user") or {}
if review.get("state") == "APPROVED" and user.get("login") == owner_login and user.get("id") == owner_id:
if author == owner_login or (pull_request.get("author_id") == owner_id):
raise MergeGateDenied("pr_author_self_approval_denied")
owner_review = review
if owner_review is None:
raise MergeGateDenied("owner_review_required")
if unresolved_conversations != 0:
raise MergeGateDenied("unresolved_conversations_block_merge")
missing = [context for context in required_contexts if statuses.get(context) != "success"]
if missing:
raise MergeGateDenied("required_checks_missing:" + ",".join(missing))
return MergeGateDecision(True, int(owner_review.get("id", 0)), tuple(required_contexts))
+167
View File
@@ -0,0 +1,167 @@
"""Fail-closed operation authorization for an isolated Git writer."""
from __future__ import annotations
import ipaddress
import os
import re
import subprocess
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
from isolated_writer_token_registry import TokenRegistry
ROUTE_OPERATIONS = {
("GET", "/api/runtime-status"): "runtime.status.read",
("GET", "/api/baseline"): "baseline.read",
("GET", "/api/patches"): "patches.read",
("POST", "/api/commit-file"): "commit_file.atomic",
}
OID_RE = re.compile(r"[0-9a-f]{40}")
class AuthorizationError(PermissionError):
def __init__(self, code: str, status: int = 403):
super().__init__(code)
self.code = code
self.status = status
@dataclass(frozen=True)
class AuthorizedGrant:
record: dict[str, Any]
operation: str
use_count: int
@property
def token_id(self) -> str:
return str(self.record["token_id"])
@property
def task_id(self) -> str:
return str(self.record["task_id"])
def operation_for(method: str, path: str) -> str:
operation = ROUTE_OPERATIONS.get((method.upper(), path))
if not operation:
raise AuthorizationError("isolated_writer_operation_forbidden")
return operation
def _parse_time(value: Any) -> datetime:
try:
parsed = datetime.fromisoformat(str(value).replace("Z", "+00:00"))
except (TypeError, ValueError):
raise AuthorizationError("writer_token_time_invalid") from None
if parsed.tzinfo is None:
raise AuthorizationError("writer_token_time_invalid")
return parsed.astimezone(timezone.utc)
def _git_is_ancestor(workspace: Path, base: str, branch: str) -> bool:
result = subprocess.run(
["git", "merge-base", "--is-ancestor", base, branch],
cwd=workspace,
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
return result.returncode == 0
def _validate_grant(
grant: dict[str, Any],
*,
audience: str,
source_ip: str,
operation: str,
branch: str | None,
path: str | None,
workspace: Path,
current_time: datetime,
) -> None:
if int(grant.get("schema_version", 0)) != 2:
raise AuthorizationError("writer_token_policy_v2_required", 401)
if str(grant.get("audience") or "") != audience:
raise AuthorizationError("writer_token_audience_rejected")
try:
peer = ipaddress.ip_address(source_ip)
networks = [ipaddress.ip_network(str(value), strict=True) for value in grant.get("source_cidrs", [])]
except ValueError:
raise AuthorizationError("writer_token_source_rejected") from None
if not networks or not any(peer in network for network in networks):
raise AuthorizationError("writer_token_source_rejected")
created = _parse_time(grant.get("created_at"))
not_before = _parse_time(grant.get("not_before"))
expires = _parse_time(grant.get("expires_at"))
if expires <= not_before or (expires - created).total_seconds() > 1800:
raise AuthorizationError("writer_token_lifetime_rejected")
if current_time < not_before:
raise AuthorizationError("writer_token_not_yet_valid", 401)
if current_time >= expires:
raise AuthorizationError("writer_token_expired", 401)
if bool(grant.get("revoked")):
raise AuthorizationError("writer_token_revoked", 401)
operations = grant.get("allowed_operations")
if not isinstance(operations, list) or operation not in operations:
raise AuthorizationError("writer_token_operation_rejected")
expected_branch = str(grant.get("allowed_branch") or "")
configured_branch = os.getenv("AIOS_GITOPS_ALLOWED_BRANCH", "")
if expected_branch != configured_branch:
raise AuthorizationError("writer_token_branch_rejected")
if branch is not None and branch != expected_branch:
raise AuthorizationError("writer_token_branch_rejected")
if path is not None:
paths = grant.get("allowed_paths")
if not isinstance(paths, list) or path not in paths:
raise AuthorizationError("writer_token_path_rejected")
base = str(grant.get("approved_base_oid") or "").lower()
if not OID_RE.fullmatch(base):
raise AuthorizationError("writer_token_base_rejected")
if not _git_is_ancestor(workspace, base, expected_branch):
raise AuthorizationError("writer_token_base_not_ancestor")
def authorize(
registry: TokenRegistry,
token: str,
*,
method: str,
path: str,
source_ip: str,
workspace: Path,
branch: str | None = None,
target_path: str | None = None,
current_time: datetime | None = None,
) -> AuthorizedGrant:
if os.getenv("AIOS_GITOPS_TOKEN_POLICY_VERSION_REQUIRED", "") != "2":
raise AuthorizationError("writer_token_policy_v2_not_required")
if not token:
raise AuthorizationError("unauthorized", 401)
grant = registry.find(token)
if grant is None:
raise AuthorizationError("unauthorized", 401)
operation = operation_for(method, path)
_validate_grant(
grant,
audience=os.getenv("AIOS_GITOPS_SERVICE_AUDIENCE", ""),
source_ip=source_ip,
operation=operation,
branch=branch,
path=target_path,
workspace=workspace,
current_time=current_time or datetime.now(timezone.utc),
)
limits = grant.get("max_uses")
if not isinstance(limits, dict):
raise AuthorizationError("writer_token_use_limit_missing")
try:
used = registry.consume(str(grant["token_id"]), operation, int(limits[operation]))
except (KeyError, TypeError, ValueError):
raise AuthorizationError("writer_token_use_limit_missing") from None
except PermissionError as exc:
code = str(exc)
raise AuthorizationError(code, 401 if code == "writer_token_revoked" else 403) from None
return AuthorizedGrant(grant, operation, used)
@@ -0,0 +1,403 @@
"""Atomic, idempotent commit-file Saga for the isolated writer."""
from __future__ import annotations
import fcntl
import hashlib
import json
import os
import re
import sqlite3
import subprocess
import tempfile
import threading
from pathlib import Path
from typing import Any
import aios_isolated_writer_policy as policy
OID_RE = re.compile(r"[0-9a-f]{40}")
IDEMPOTENCY_RE = re.compile(r"[A-Za-z0-9][A-Za-z0-9._:-]{7,127}")
STATES = frozenset({
"PENDING", "FILE_WRITTEN", "COMMITTED", "PUSHED", "VERIFIED", "COMPLETED",
"FAILED_RECOVERABLE", "MANUAL_REVIEW_REQUIRED",
})
_LOCKS: dict[str, threading.Lock] = {}
_LOCKS_GUARD = threading.Lock()
_LEDGER_INIT_LOCK = threading.Lock()
class CommitFileError(RuntimeError):
def __init__(self, code: str, status: int = 400):
super().__init__(code)
self.code = code
self.status = status
class SimulatedCrash(RuntimeError):
pass
def canonical_digest(data: dict[str, Any]) -> str:
payload = {
"branch": data["branch"],
"path": data["path"],
"content_sha256": hashlib.sha256(data["content"].encode("utf-8")).hexdigest(),
"message": data["message"],
"expected_head_oid": data["expected_head_oid"],
}
encoded = json.dumps(payload, sort_keys=True, separators=(",", ":")).encode("utf-8")
return hashlib.sha256(encoded).hexdigest()
class OperationLedger:
def __init__(self, path: Path | str):
self.path = Path(path)
self.path.parent.mkdir(parents=True, exist_ok=True)
if self.path.parent.is_symlink() or self.path.is_symlink():
raise PermissionError("writer_operation_ledger_symlink_rejected")
with _LEDGER_INIT_LOCK:
with self.connect() as db:
db.execute(
"""
CREATE TABLE IF NOT EXISTS writer_operations (
operation_id TEXT PRIMARY KEY,
token_id TEXT NOT NULL,
task_id TEXT NOT NULL,
idempotency_key TEXT NOT NULL UNIQUE,
operation_digest TEXT NOT NULL,
branch TEXT NOT NULL,
path TEXT NOT NULL,
content_sha256 TEXT NOT NULL,
pre_head_oid TEXT NOT NULL,
commit_oid TEXT,
push_status TEXT,
verification_status TEXT,
state TEXT NOT NULL,
created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
error_code TEXT
)
"""
)
os.chmod(self.path, 0o600)
def connect(self) -> sqlite3.Connection:
db = sqlite3.connect(self.path, timeout=30)
db.row_factory = sqlite3.Row
db.execute("PRAGMA journal_mode=WAL")
db.execute("PRAGMA synchronous=FULL")
return db
def get_or_create(self, grant: Any, data: dict[str, Any], digest: str) -> dict[str, Any]:
key = data["idempotency_key"]
with self.connect() as db:
db.execute("BEGIN IMMEDIATE")
row = db.execute(
"SELECT * FROM writer_operations WHERE idempotency_key=?", (key,)
).fetchone()
if row:
db.commit()
record = dict(row)
if record["operation_digest"] != digest:
raise CommitFileError("writer_idempotency_conflict", 409)
if (
record["token_id"] != grant.token_id
or record["task_id"] != grant.task_id
):
raise CommitFileError("writer_idempotency_owner_conflict", 409)
return record
operation_id = hashlib.sha256(
f"{grant.task_id}\0{key}".encode("utf-8")
).hexdigest()
content_sha = hashlib.sha256(data["content"].encode("utf-8")).hexdigest()
db.execute(
"""
INSERT INTO writer_operations(
operation_id,token_id,task_id,idempotency_key,operation_digest,
branch,path,content_sha256,pre_head_oid,state
) VALUES(?,?,?,?,?,?,?,?,?,?)
""",
(
operation_id, grant.token_id, grant.task_id, key, digest,
data["branch"], data["path"], content_sha,
data["expected_head_oid"], "PENDING",
),
)
db.commit()
return self.get(key)
def get(self, key: str) -> dict[str, Any]:
with self.connect() as db:
row = db.execute(
"SELECT * FROM writer_operations WHERE idempotency_key=?", (key,)
).fetchone()
if row is None:
raise KeyError(key)
return dict(row)
def update(self, key: str, state: str, **fields: Any) -> dict[str, Any]:
if state not in STATES:
raise ValueError("invalid writer operation state")
allowed = {"commit_oid", "push_status", "verification_status", "error_code"}
assignments = ["state=?", "updated_at=CURRENT_TIMESTAMP"]
values: list[Any] = [state]
for name, value in fields.items():
if name not in allowed:
raise ValueError("invalid ledger field")
assignments.append(f"{name}=?")
values.append(value)
values.append(key)
with self.connect() as db:
db.execute("BEGIN IMMEDIATE")
db.execute(
f"UPDATE writer_operations SET {','.join(assignments)} WHERE idempotency_key=?",
values,
)
db.commit()
return self.get(key)
class AtomicCommitFile:
def __init__(
self,
*,
workspace: Path,
bare_repo: Path,
ledger: OperationLedger,
remote: str,
audit,
crash_after_state: str | None = None,
):
self.workspace = workspace
self.bare_repo = bare_repo
self.ledger = ledger
self.remote = remote
self.audit = audit
self.crash_after_state = crash_after_state
def _git(self, *args: str, check: bool = True) -> str:
result = subprocess.run(
["git", *args], cwd=self.workspace, text=True, capture_output=True
)
if check and result.returncode != 0:
raise CommitFileError("writer_git_operation_failed", 500)
return result.stdout.strip()
def _transition(self, key: str, state: str, **fields: Any) -> dict[str, Any]:
record = self.ledger.update(key, state, **fields)
if self.crash_after_state == state:
raise SimulatedCrash(state)
return record
def _validate(self, data: Any) -> dict[str, Any]:
if not isinstance(data, dict):
raise CommitFileError("writer_request_object_required")
required = ("branch", "path", "content", "message", "expected_head_oid", "idempotency_key")
if any(field not in data for field in required):
raise CommitFileError("writer_request_incomplete")
if data.get("force") or data.get("force_push"):
raise CommitFileError("isolated_writer_force_rejected", 403)
if set(data) != set(required):
raise CommitFileError("writer_request_fields_rejected")
item = {field: data[field] for field in required}
if item["branch"] != policy.allowed_branch():
raise CommitFileError("isolated_writer_branch_rejected", 403)
try:
item["path"] = policy.validate_path(item["path"], self.workspace)
except PermissionError as exc:
raise CommitFileError(str(exc), 403) from None
if not isinstance(item["content"], str) or "\x00" in item["content"]:
raise CommitFileError("writer_content_rejected")
try:
policy.validate_content_for_path(item["path"], item["content"])
except PermissionError as exc:
raise CommitFileError(str(exc), 403) from None
if len(item["content"].encode("utf-8")) > 1_048_576:
raise CommitFileError("writer_content_too_large", 413)
message = item["message"]
if (
not isinstance(message, str)
or not 1 <= len(message) <= 200
or "\x00" in message
or "\n" in message
or "\r" in message
):
raise CommitFileError("writer_message_rejected")
if not isinstance(item["expected_head_oid"], str) or not OID_RE.fullmatch(item["expected_head_oid"]):
raise CommitFileError("writer_expected_head_rejected")
if not isinstance(item["idempotency_key"], str) or not IDEMPOTENCY_RE.fullmatch(item["idempotency_key"]):
raise CommitFileError("writer_idempotency_key_rejected")
return item
def _atomic_write(self, relative: str, content: str) -> None:
target = self.workspace / relative
target.parent.mkdir(parents=True, exist_ok=True)
fd, temporary = tempfile.mkstemp(prefix=".writer-", dir=target.parent)
try:
with os.fdopen(fd, "wb", closefd=True) as stream:
stream.write(content.encode("utf-8"))
stream.flush()
os.fsync(stream.fileno())
os.replace(temporary, target)
parent_fd = os.open(target.parent, os.O_RDONLY | getattr(os, "O_DIRECTORY", 0))
try:
os.fsync(parent_fd)
finally:
os.close(parent_fd)
finally:
if os.path.exists(temporary):
os.unlink(temporary)
def _verify_remote(self, branch: str, oid: str) -> None:
bare = subprocess.run(
["git", "--git-dir", str(self.bare_repo), "rev-parse", f"refs/heads/{branch}"],
text=True, capture_output=True,
)
if bare.returncode != 0 or bare.stdout.strip() != oid:
raise CommitFileError("writer_authoritative_bare_oid_mismatch", 500)
remote = subprocess.run(
["git", "ls-remote", "--heads", self.remote, f"refs/heads/{branch}"],
cwd=self.workspace, text=True, capture_output=True,
)
if remote.returncode != 0:
raise CommitFileError("writer_gitea_verification_failed", 503)
fields = remote.stdout.strip().split()
if len(fields) != 2:
raise CommitFileError("writer_gitea_ref_missing", 503)
if fields[0] != oid:
raise CommitFileError("writer_gitea_oid_conflict", 409)
def execute(self, grant: Any, raw_data: Any) -> dict[str, Any]:
data = self._validate(raw_data)
digest = canonical_digest(data)
branch = data["branch"]
with _LOCKS_GUARD:
lock = _LOCKS.setdefault(branch, threading.Lock())
lock_path = self.ledger.path.parent / f"{hashlib.sha256(branch.encode()).hexdigest()}.lock"
with lock, open(lock_path, "a+b") as branch_lock:
fcntl.flock(branch_lock.fileno(), fcntl.LOCK_EX)
return self._execute_locked(grant, data, digest)
def _execute_locked(self, grant: Any, data: dict[str, Any], digest: str) -> dict[str, Any]:
key = data["idempotency_key"]
record = self.ledger.get_or_create(grant, data, digest)
if record["state"] == "COMPLETED":
return self._result(record, idempotent=True)
if self._git("rev-parse", "--abbrev-ref", "HEAD") != data["branch"]:
record = self._transition(
key, "MANUAL_REVIEW_REQUIRED", error_code="writer_current_branch_conflict"
)
self._audit(grant, data, digest, record, "manual_review_required")
raise CommitFileError("writer_current_branch_conflict", 409)
head = self._git("rev-parse", "HEAD")
commit_oid = record.get("commit_oid")
if head != record["pre_head_oid"] and head != commit_oid:
record = self._transition(
key, "MANUAL_REVIEW_REQUIRED", error_code="writer_branch_head_conflict"
)
self._audit(grant, data, digest, record, "manual_review_required")
raise CommitFileError("writer_branch_head_conflict", 409)
try:
state = record["state"]
if state == "MANUAL_REVIEW_REQUIRED":
raise CommitFileError("writer_manual_review_required", 409)
if state == "FAILED_RECOVERABLE":
if record.get("push_status") == "confirmed":
state = "PUSHED"
else:
state = "COMMITTED" if commit_oid else "FILE_WRITTEN"
if state == "PENDING":
self._atomic_write(data["path"], data["content"])
record = self._transition(key, "FILE_WRITTEN")
state = record["state"]
if state == "FILE_WRITTEN":
actual_sha = hashlib.sha256(
(self.workspace / data["path"]).read_bytes()
).hexdigest()
if actual_sha != record["content_sha256"]:
raise CommitFileError("writer_file_content_conflict", 409)
staged_before = self._git("diff", "--cached", "--name-only", check=False).splitlines()
if staged_before:
raise CommitFileError("writer_extra_staged_files_rejected", 409)
self._git("add", "--", data["path"])
staged = self._git("diff", "--cached", "--name-only").splitlines()
if staged != [data["path"]]:
raise CommitFileError("writer_extra_staged_files_rejected", 409)
self._git("commit", "-m", data["message"])
commit_oid = self._git("rev-parse", "HEAD")
record = self._transition(key, "COMMITTED", commit_oid=commit_oid)
state = record["state"]
if state == "COMMITTED":
commit_oid = str(record["commit_oid"])
pushed = subprocess.run(
["git", "push", "--porcelain", self.remote, f"HEAD:refs/heads/{data['branch']}"],
cwd=self.workspace, text=True, capture_output=True,
)
if pushed.returncode != 0:
self._transition(
key, "FAILED_RECOVERABLE",
push_status="failed", error_code="writer_gitea_push_failed",
)
raise CommitFileError("writer_gitea_push_failed", 503)
record = self._transition(key, "PUSHED", push_status="confirmed")
state = record["state"]
if state == "PUSHED":
commit_oid = str(record["commit_oid"])
self._verify_remote(data["branch"], commit_oid)
record = self._transition(
key, "VERIFIED", verification_status="bare+gitea"
)
state = record["state"]
if state == "VERIFIED":
record = self._transition(key, "COMPLETED", error_code=None)
self._audit(grant, data, digest, record, "completed")
return self._result(record, idempotent=False)
except SimulatedCrash:
raise
except CommitFileError as exc:
latest = self.ledger.get(key)
if latest["state"] not in {"FAILED_RECOVERABLE", "MANUAL_REVIEW_REQUIRED"}:
failure_state = (
"MANUAL_REVIEW_REQUIRED"
if exc.code in {
"writer_gitea_oid_conflict",
"writer_file_content_conflict",
"writer_current_branch_conflict",
}
else "FAILED_RECOVERABLE"
)
latest = self._transition(key, failure_state, error_code=exc.code)
self._audit(grant, data, digest, latest, "failed")
raise
def _audit(self, grant: Any, data: dict[str, Any], digest: str, record: dict[str, Any], result: str) -> None:
self.audit({
"token_id": grant.token_id,
"task_id": grant.task_id,
"audience": grant.record["audience"],
"source_ip": grant.record.get("_source_ip", "socket-peer"),
"operation": grant.operation,
"branch": data["branch"],
"path": data["path"],
"idempotency_key": data["idempotency_key"],
"operation_digest": digest,
"path_policy_sha256": policy.configured_path_policy_sha256(),
"pre_head_oid": record["pre_head_oid"],
"commit_oid": record.get("commit_oid"),
"result": result,
})
@staticmethod
def _result(record: dict[str, Any], *, idempotent: bool) -> dict[str, Any]:
return {
"status": "completed",
"state": record["state"],
"operation_id": record["operation_id"],
"branch": record["branch"],
"path": record["path"],
"commit_oid": record.get("commit_oid"),
"push_status": record.get("push_status"),
"verification_status": record.get("verification_status"),
"idempotent": idempotent,
}
@@ -0,0 +1,422 @@
"""Strict service-level exact-path policy for isolated Git writers."""
from __future__ import annotations
import grp
import hashlib
import json
import os
import re
import stat
from dataclasses import dataclass
from functools import lru_cache
from pathlib import Path, PurePosixPath, PureWindowsPath
from typing import Any
from urllib.parse import unquote
MAX_POLICY_BYTES = 65_536
POLICY_FIELDS = frozenset(
{
"schema_version",
"policy_id",
"audience",
"repository_identity",
"allowed_branch",
"allowed_operations",
}
)
OPERATION_FIELDS = frozenset({"exact_paths", "max_content_bytes", "content_profile"})
SUPPORTED_OPERATIONS = frozenset({"commit_file.atomic"})
SUPPORTED_CONTENT_PROFILES = frozenset({"systemd-codex-review-worker-v1"})
_SECRET_RE = re.compile(
r"(?i)(authorization\s*:|bearer\s+|(?:token|password|secret)\s*[:=])"
)
_SYSTEMD_NAME_RE = re.compile(r"[A-Za-z][A-Za-z0-9]*\Z")
_SYSTEMD_MAX_LINE_BYTES = 4096
_SYSTEMD_ALLOWED_DIRECTIVES = {
"Unit": frozenset({"Description", "After", "Wants"}),
"Service": frozenset(
{
"Type",
"User",
"Group",
"WorkingDirectory",
"Environment",
"ExecStart",
"Restart",
"RestartSec",
"NoNewPrivileges",
"PrivateTmp",
"PrivateDevices",
"ProtectSystem",
"ProtectHome",
"ProtectKernelTunables",
"ProtectKernelModules",
"ProtectControlGroups",
"LockPersonality",
"RestrictSUIDSGID",
"RestrictRealtime",
"MemoryDenyWriteExecute",
"RuntimeDirectory",
"StateDirectory",
"CacheDirectory",
"ReadOnlyPaths",
"ReadWritePaths",
"UMask",
}
),
"Install": frozenset({"WantedBy"}),
}
_SYSTEMD_MULTI_DIRECTIVES = frozenset(
{
("Service", "Environment"),
("Service", "ReadOnlyPaths"),
("Service", "ReadWritePaths"),
}
)
_SYSTEMD_EXACT_SINGLE_VALUES = {
("Unit", "Description"): "AIOS Codex Review/Test Worker",
("Unit", "After"): "network-online.target",
("Unit", "Wants"): "network-online.target",
("Service", "Type"): "simple",
("Service", "User"): "aiman",
("Service", "Group"): "gitops",
(
"Service",
"WorkingDirectory",
): "/usr/libexec/aios-codex-review-test-worker/current",
(
"Service",
"ExecStart",
): "/usr/bin/python3 -m task_plane.workers.codex_review_test",
("Service", "Restart"): "on-failure",
("Service", "RestartSec"): "2",
("Service", "NoNewPrivileges"): "true",
("Service", "PrivateTmp"): "true",
("Service", "PrivateDevices"): "true",
("Service", "ProtectSystem"): "strict",
("Service", "ProtectHome"): "true",
("Service", "ProtectKernelTunables"): "true",
("Service", "ProtectKernelModules"): "true",
("Service", "ProtectControlGroups"): "true",
("Service", "LockPersonality"): "true",
("Service", "RestrictSUIDSGID"): "true",
("Service", "RestrictRealtime"): "true",
("Service", "MemoryDenyWriteExecute"): "true",
("Service", "RuntimeDirectory"): "aios-codex-review-test-worker",
("Service", "StateDirectory"): "aios-codex-review-test-worker",
("Service", "CacheDirectory"): "aios-codex-review-test-worker",
("Service", "UMask"): "0077",
("Install", "WantedBy"): "multi-user.target",
}
_SYSTEMD_EXACT_MULTI_VALUES = {
("Service", "Environment"): (
"AIOS_CODEX_WORKER_MODE=shadow",
"AIOS_TASK_PLANE_EXECUTION_ENABLED=false",
"PYTHONDONTWRITEBYTECODE=1",
),
("Service", "ReadOnlyPaths"): (
"/data/aios-gitops/repo.git",
"/usr/libexec/aios-codex-review-test-worker",
),
("Service", "ReadWritePaths"): (
"/run/aios-codex-review-test-worker",
"/var/lib/aios-codex-review-test-worker",
"/var/cache/aios-codex-review-test-worker",
),
}
class PathPolicyError(PermissionError):
pass
@dataclass(frozen=True)
class OperationPolicy:
exact_paths: tuple[str, ...]
max_content_bytes: int
content_profile: str
@dataclass(frozen=True)
class ServicePathPolicy:
schema_version: int
policy_id: str
audience: str
repository_identity: str
allowed_branch: str
allowed_operations: dict[str, OperationPolicy]
sha256: str
def operation(self, name: str) -> OperationPolicy:
try:
return self.allowed_operations[name]
except KeyError:
raise PathPolicyError("isolated_writer_service_operation_rejected") from None
def require_exact_path(self, operation: str, path: str) -> OperationPolicy:
selected = self.operation(operation)
if path not in selected.exact_paths:
raise PathPolicyError("isolated_writer_service_path_rejected")
return selected
def _canonical_json(value: Any) -> bytes:
return (
json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+ "\n"
).encode("utf-8")
def validate_relative_path_syntax(path: str) -> str:
if not isinstance(path, str):
raise PathPolicyError("isolated_writer_path_rejected")
if not path or path != path.strip() or "\x00" in path or "\\" in path:
raise PathPolicyError("isolated_writer_path_rejected")
try:
decoded = unquote(path, errors="strict")
except (UnicodeDecodeError, ValueError):
raise PathPolicyError("isolated_writer_path_rejected") from None
if decoded != path:
raise PathPolicyError("isolated_writer_path_rejected")
posix = PurePosixPath(path)
windows = PureWindowsPath(path)
parts = path.split("/")
if (
posix.is_absolute()
or windows.is_absolute()
or bool(windows.drive)
or any(part in {"", ".", ".."} for part in parts)
):
raise PathPolicyError("isolated_writer_path_rejected")
return path
def _require_fields(value: dict[str, Any], allowed: frozenset[str], code: str) -> None:
if set(value) != allowed:
raise PathPolicyError(code)
def _parse_policy(document: Any, canonical: bytes) -> ServicePathPolicy:
if not isinstance(document, dict):
raise PathPolicyError("isolated_writer_path_policy_object_required")
_require_fields(document, POLICY_FIELDS, "isolated_writer_path_policy_fields_rejected")
if document["schema_version"] != 1:
raise PathPolicyError("isolated_writer_path_policy_schema_rejected")
text_fields = (
"policy_id",
"audience",
"repository_identity",
"allowed_branch",
)
if any(
not isinstance(document[field], str) or not document[field].strip()
for field in text_fields
):
raise PathPolicyError("isolated_writer_path_policy_value_rejected")
raw_operations = document["allowed_operations"]
if not isinstance(raw_operations, dict) or not raw_operations:
raise PathPolicyError("isolated_writer_path_policy_operations_rejected")
if not set(raw_operations).issubset(SUPPORTED_OPERATIONS):
raise PathPolicyError("isolated_writer_path_policy_operations_rejected")
operations: dict[str, OperationPolicy] = {}
for name, raw in raw_operations.items():
if not isinstance(raw, dict):
raise PathPolicyError("isolated_writer_path_policy_operation_rejected")
_require_fields(
raw,
OPERATION_FIELDS,
"isolated_writer_path_policy_operation_fields_rejected",
)
paths = raw["exact_paths"]
if (
not isinstance(paths, list)
or not paths
or any(not isinstance(item, str) for item in paths)
):
raise PathPolicyError("isolated_writer_path_policy_paths_rejected")
validated = tuple(validate_relative_path_syntax(item) for item in paths)
if len(set(validated)) != len(validated):
raise PathPolicyError("isolated_writer_path_policy_paths_rejected")
maximum = raw["max_content_bytes"]
if not isinstance(maximum, int) or isinstance(maximum, bool) or not 1 <= maximum <= 1_048_576:
raise PathPolicyError("isolated_writer_path_policy_size_rejected")
profile = raw["content_profile"]
if profile not in SUPPORTED_CONTENT_PROFILES:
raise PathPolicyError("isolated_writer_path_policy_profile_rejected")
operations[name] = OperationPolicy(validated, maximum, profile)
return ServicePathPolicy(
schema_version=1,
policy_id=document["policy_id"],
audience=document["audience"],
repository_identity=document["repository_identity"],
allowed_branch=document["allowed_branch"],
allowed_operations=operations,
sha256=hashlib.sha256(canonical).hexdigest(),
)
def load_path_policy(
path: Path,
*,
expected_owner_uid: int = 0,
expected_group_gid: int | None = None,
expected_parent_mode: int = 0o750,
expected_file_mode: int = 0o640,
) -> ServicePathPolicy:
lexical = Path(os.path.abspath(str(path)))
if lexical.is_symlink():
raise PathPolicyError("isolated_writer_path_policy_symlink_rejected")
try:
parent_info = lexical.parent.stat(follow_symlinks=False)
file_info = lexical.stat(follow_symlinks=False)
except FileNotFoundError:
raise PathPolicyError("isolated_writer_path_policy_missing") from None
if stat.S_ISLNK(parent_info.st_mode) or not stat.S_ISDIR(parent_info.st_mode):
raise PathPolicyError("isolated_writer_path_policy_parent_rejected")
if stat.S_ISLNK(file_info.st_mode) or not stat.S_ISREG(file_info.st_mode):
raise PathPolicyError("isolated_writer_path_policy_file_rejected")
if lexical.resolve(strict=True) != lexical:
raise PathPolicyError("isolated_writer_path_policy_symlink_rejected")
if (
parent_info.st_uid != expected_owner_uid
or stat.S_IMODE(parent_info.st_mode) != expected_parent_mode
):
raise PathPolicyError("isolated_writer_path_policy_parent_permissions_rejected")
if (
file_info.st_uid != expected_owner_uid
or stat.S_IMODE(file_info.st_mode) != expected_file_mode
):
raise PathPolicyError("isolated_writer_path_policy_permissions_rejected")
if expected_group_gid is not None and (
parent_info.st_gid != expected_group_gid or file_info.st_gid != expected_group_gid
):
raise PathPolicyError("isolated_writer_path_policy_group_rejected")
if file_info.st_size <= 0 or file_info.st_size > MAX_POLICY_BYTES:
raise PathPolicyError("isolated_writer_path_policy_file_size_rejected")
raw = lexical.read_bytes()
try:
document = json.loads(raw.decode("utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError):
raise PathPolicyError("isolated_writer_path_policy_json_rejected") from None
canonical = _canonical_json(document)
if raw != canonical:
raise PathPolicyError("isolated_writer_path_policy_not_canonical")
return _parse_policy(document, canonical)
@lru_cache(maxsize=1)
def configured_path_policy() -> ServicePathPolicy | None:
configured = os.getenv("AIOS_GITOPS_PATH_POLICY_FILE", "").strip()
if not configured:
return None
if not Path(configured).is_absolute():
raise PathPolicyError("isolated_writer_path_policy_path_rejected")
try:
gitops_gid = grp.getgrnam("gitops").gr_gid
except KeyError:
raise PathPolicyError("isolated_writer_path_policy_group_rejected") from None
loaded = load_path_policy(Path(configured), expected_group_gid=gitops_gid)
if loaded.audience != os.getenv("AIOS_GITOPS_SERVICE_AUDIENCE", ""):
raise PathPolicyError("isolated_writer_path_policy_audience_rejected")
if loaded.repository_identity != os.getenv("AIOS_GITOPS_REPOSITORY_IDENTITY", ""):
raise PathPolicyError("isolated_writer_path_policy_repository_rejected")
if loaded.allowed_branch != os.getenv("AIOS_GITOPS_ALLOWED_BRANCH", ""):
raise PathPolicyError("isolated_writer_path_policy_branch_rejected")
return loaded
def clear_path_policy_cache() -> None:
configured_path_policy.cache_clear()
def require_service_path(operation: str, path: str) -> OperationPolicy:
loaded = configured_path_policy()
if loaded is None:
raise PathPolicyError("isolated_writer_service_path_policy_required")
return loaded.require_exact_path(operation, path)
def configured_path_policy_sha256() -> str | None:
loaded = configured_path_policy()
return None if loaded is None else loaded.sha256
def validate_content(operation: str, path: str, content: str) -> None:
selected = require_service_path(operation, path)
try:
encoded = content.encode("utf-8", errors="strict")
except (AttributeError, UnicodeEncodeError):
raise PathPolicyError("isolated_writer_systemd_profile_rejected") from None
if len(encoded) > selected.max_content_bytes:
raise PathPolicyError("isolated_writer_service_content_too_large")
if selected.content_profile == "systemd-codex-review-worker-v1":
_validate_systemd_codex_review_worker(content)
return
raise PathPolicyError("isolated_writer_path_policy_profile_rejected")
def _validate_systemd_codex_review_worker(content: str) -> None:
if not isinstance(content, str):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
try:
content.encode("utf-8", errors="strict")
except UnicodeEncodeError:
raise PathPolicyError("isolated_writer_systemd_profile_rejected") from None
if (
"\x00" in content
or "\r" in content
or _SECRET_RE.search(content)
or not content.endswith("\n")
):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
sections: set[str] = set()
values: dict[tuple[str, str], list[str]] = {}
section: str | None = None
for line in content.splitlines():
if len(line.encode("utf-8")) > _SYSTEMD_MAX_LINE_BYTES:
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
if not line:
continue
if line[0].isspace() or line.startswith(("#", ";")):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
if line.startswith("[") and line.endswith("]"):
candidate = line[1:-1]
if (
not _SYSTEMD_NAME_RE.fullmatch(candidate)
or candidate not in _SYSTEMD_ALLOWED_DIRECTIVES
or candidate in sections
):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
sections.add(candidate)
section = candidate
continue
if section is None or "=" not in line:
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
directive, value = line.split("=", 1)
if (
not _SYSTEMD_NAME_RE.fullmatch(directive)
or directive not in _SYSTEMD_ALLOWED_DIRECTIVES[section]
or not value
or value != value.strip()
):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
key = (section, directive)
if key in values and key not in _SYSTEMD_MULTI_DIRECTIVES:
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
values.setdefault(key, []).append(value)
if sections != set(_SYSTEMD_ALLOWED_DIRECTIVES):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
if set(values) != set(_SYSTEMD_EXACT_SINGLE_VALUES) | set(
_SYSTEMD_EXACT_MULTI_VALUES
):
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
for key, expected in _SYSTEMD_EXACT_SINGLE_VALUES.items():
if values.get(key) != [expected]:
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
for key, expected in _SYSTEMD_EXACT_MULTI_VALUES.items():
observed = values.get(key, [])
if len(observed) != len(set(observed)) or tuple(observed) != expected:
raise PathPolicyError("isolated_writer_systemd_profile_rejected")
@@ -0,0 +1,241 @@
"""Digest-only, task-scoped token registry for isolated writers."""
from __future__ import annotations
import argparse
import hashlib
import hmac
import json
import os
import grp
import pwd
import secrets
import sqlite3
import stat
import threading
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
_REGISTRY_INIT_LOCK = threading.Lock()
def utc_now() -> str:
return datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
def token_digest(token: str) -> str:
return hashlib.sha256(token.encode("utf-8")).hexdigest()
class TokenRegistry:
def __init__(self, path: Path | str):
self.path = Path(path)
self._prepare()
def _prepare(self) -> None:
self.path.parent.mkdir(parents=True, exist_ok=True)
if self.path.parent.is_symlink() or self.path.is_symlink():
raise PermissionError("writer_token_registry_symlink_rejected")
with _REGISTRY_INIT_LOCK:
with self._connect() as db:
db.executescript(
"""
CREATE TABLE IF NOT EXISTS token_grants (
token_id TEXT PRIMARY KEY,
token_digest TEXT NOT NULL UNIQUE,
grant_json TEXT NOT NULL,
revoked INTEGER NOT NULL DEFAULT 0,
created_at TEXT NOT NULL
);
CREATE TABLE IF NOT EXISTS token_uses (
token_id TEXT NOT NULL,
operation TEXT NOT NULL,
use_count INTEGER NOT NULL DEFAULT 0,
PRIMARY KEY(token_id, operation),
FOREIGN KEY(token_id) REFERENCES token_grants(token_id)
);
"""
)
os.chmod(self.path, 0o600)
def _connect(self) -> sqlite3.Connection:
db = sqlite3.connect(self.path, timeout=30, isolation_level=None)
db.row_factory = sqlite3.Row
db.execute("PRAGMA journal_mode=WAL")
db.execute("PRAGMA synchronous=FULL")
db.execute("PRAGMA foreign_keys=ON")
return db
def issue(self, grant: dict[str, Any], token: str | None = None) -> tuple[str, dict[str, Any]]:
value = token or secrets.token_urlsafe(32)
if len(value.encode("utf-8")) < 32:
raise ValueError("writer_token_entropy_insufficient")
record = dict(grant)
record["schema_version"] = 2
record["token_digest"] = token_digest(value)
record["revoked"] = False
record.setdefault("created_at", utc_now())
required = (
"token_id",
"audience",
"task_id",
"source_cidrs",
"allowed_branch",
"approved_base_oid",
"allowed_operations",
"allowed_paths",
"not_before",
"expires_at",
"max_uses",
"issued_by",
)
if any(not record.get(field) for field in required):
raise ValueError("writer_token_grant_incomplete")
encoded = json.dumps(record, sort_keys=True, separators=(",", ":"))
with self._connect() as db:
db.execute("BEGIN IMMEDIATE")
try:
db.execute(
"""
INSERT INTO token_grants(token_id, token_digest, grant_json, revoked, created_at)
VALUES(?,?,?,?,?)
""",
(
str(record["token_id"]),
str(record["token_digest"]),
encoded,
0,
str(record["created_at"]),
),
)
db.execute("COMMIT")
except Exception:
db.execute("ROLLBACK")
raise
return value, record
def find(self, token: str) -> dict[str, Any] | None:
candidate = token_digest(token)
with self._connect() as db:
rows = db.execute(
"SELECT token_digest, grant_json, revoked FROM token_grants"
).fetchall()
match = None
for row in rows:
if hmac.compare_digest(str(row["token_digest"]), candidate):
match = dict(json.loads(str(row["grant_json"])))
match["revoked"] = bool(row["revoked"])
return match
def consume(self, token_id: str, operation: str, maximum: int) -> int:
if maximum < 1:
raise PermissionError("writer_token_operation_exhausted")
with self._connect() as db:
db.execute("BEGIN IMMEDIATE")
try:
row = db.execute(
"SELECT revoked FROM token_grants WHERE token_id=?",
(token_id,),
).fetchone()
if row is None or bool(row["revoked"]):
raise PermissionError("writer_token_revoked")
used_row = db.execute(
"SELECT use_count FROM token_uses WHERE token_id=? AND operation=?",
(token_id, operation),
).fetchone()
used = int(used_row["use_count"]) if used_row else 0
if used >= maximum:
raise PermissionError("writer_token_operation_exhausted")
current = used + 1
db.execute(
"""
INSERT INTO token_uses(token_id, operation, use_count)
VALUES(?,?,?)
ON CONFLICT(token_id, operation)
DO UPDATE SET use_count=excluded.use_count
""",
(token_id, operation, current),
)
db.execute("COMMIT")
return current
except Exception:
db.execute("ROLLBACK")
raise
def revoke(self, token_id: str) -> bool:
with self._connect() as db:
db.execute("BEGIN IMMEDIATE")
changed = db.execute(
"UPDATE token_grants SET revoked=1 WHERE token_id=? AND revoked=0",
(token_id,),
).rowcount
db.execute("COMMIT")
return bool(changed)
def usage(self, token_id: str, operation: str) -> int:
with self._connect() as db:
row = db.execute(
"SELECT use_count FROM token_uses WHERE token_id=? AND operation=?",
(token_id, operation),
).fetchone()
return int(row["use_count"]) if row else 0
def _write_secret_file(
path: Path,
token: str,
*,
owner: str | None = None,
group: str | None = None,
) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
if path.parent.is_symlink() or path.is_symlink():
raise PermissionError("writer_token_file_symlink_rejected")
flags = os.O_WRONLY | os.O_CREAT | os.O_EXCL
if hasattr(os, "O_NOFOLLOW"):
flags |= os.O_NOFOLLOW
fd = os.open(path, flags, 0o600)
try:
if not stat.S_ISREG(os.fstat(fd).st_mode):
raise PermissionError("writer_token_file_not_regular")
if owner is not None or group is not None:
uid = pwd.getpwnam(owner).pw_uid if owner is not None else -1
gid = grp.getgrnam(group).gr_gid if group is not None else -1
os.fchown(fd, uid, gid)
os.write(fd, token.encode("utf-8"))
os.fsync(fd)
finally:
os.close(fd)
def main(argv: list[str] | None = None) -> int:
parser = argparse.ArgumentParser()
parser.add_argument("--registry", required=True)
sub = parser.add_subparsers(dest="command", required=True)
issue = sub.add_parser("issue")
issue.add_argument("--grant-file", required=True)
issue.add_argument("--token-file", required=True)
issue.add_argument("--owner")
issue.add_argument("--group")
revoke = sub.add_parser("revoke")
revoke.add_argument("--token-id", required=True)
args = parser.parse_args(argv)
registry = TokenRegistry(args.registry)
if args.command == "issue":
grant = json.loads(Path(args.grant_file).read_text(encoding="utf-8"))
token, record = registry.issue(grant)
_write_secret_file(
Path(args.token_file),
token,
owner=args.owner,
group=args.group,
)
print(json.dumps({"token_id": record["token_id"], "created": True}, sort_keys=True))
return 0
changed = registry.revoke(args.token_id)
print(json.dumps({"token_id": args.token_id, "revoked": changed}, sort_keys=True))
return 0
if __name__ == "__main__":
raise SystemExit(main())
@@ -0,0 +1,611 @@
"""Audited Owner authorization broker for task-scoped writer provisioning.
The broker closes the bootstrap gap without minting a new general-purpose
Owner token and without modifying the bearer-token registry. It accepts an
already-authenticated, explicitly allow-listed Owner principal, records a
tamper-evident authorization event, and invokes the existing
AuthorizedTaskWriterProvisioner with a generated authorization reference.
The broker never writes Git refs itself. The governed signer, validator and
one-shot provisioner remain the only components allowed to create the task
branch/worktree/writer.
"""
from __future__ import annotations
import fcntl
import hashlib
import json
import os
import re
import tempfile
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Mapping, Protocol
from authorized_task_writer_provisioning import (
AuthorizedTaskWriterProvisioner,
BRANCH_RE,
OID_RE,
OwnerContext,
ProvisioningError,
REQUIRED_ALLOWED_OPERATIONS,
REQUIRED_DENIED_OPERATIONS,
TASK_RE,
)
ACTION = "AUTHORIZE_TASK_WRITER_PROVISIONING"
SCHEMA = "aios.owner-authorization-event.v1"
BROKER_REQUEST_FIELDS = frozenset(
{
"action",
"task_id",
"repository",
"base_ref",
"base_oid",
"target_branch",
"one_time_use",
"expires_in_seconds",
"production_change_authorized",
"tag_change_authorized",
"baseline_change_authorized",
"main_write_authorized",
"allowed_operations",
"denied_operations",
}
)
IDENTITY_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._@:+/-]{1,254}$")
SCOPE_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:-]{1,127}$")
CREDENTIAL_ID_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:@+-]{1,255}$")
WILDCARD_SCOPE_RE = re.compile(r"(^\*$)|(^.+:\*$)")
class OwnerAuthorizationError(RuntimeError):
"""Stable fail-closed broker error."""
def __init__(
self,
code: str,
status: int = 400,
*,
details: Mapping[str, Any] | None = None,
) -> None:
super().__init__(code)
self.code = code
self.status = status
self.details = dict(details or {})
@dataclass(frozen=True)
class OwnerPrincipal:
"""Non-secret authentication evidence returned by the existing API."""
identity: str
role: str
authenticated_scope: str
token_scopes: frozenset[str]
credential_id: str = ""
@dataclass(frozen=True)
class OwnerAuthorizationBrokerConfig:
repository_identity: str
state_root: Path
allowed_owner_identities: frozenset[str]
allowed_intake_scopes: frozenset[str]
allowed_credential_ids: frozenset[str] = frozenset()
max_expiry_seconds: int = 900
@classmethod
def from_environment(cls) -> "OwnerAuthorizationBrokerConfig":
identities = _csv_set(
os.getenv("AIOS_GITOPS_OWNER_AUTHORIZATION_ALLOWED_IDENTITIES", "")
)
scopes = _csv_set(
os.getenv("AIOS_GITOPS_OWNER_AUTHORIZATION_INTAKE_SCOPES", "")
)
credential_ids = _csv_set(
os.getenv("AIOS_GITOPS_OWNER_AUTHORIZATION_ALLOWED_CREDENTIAL_IDS", "")
)
if not identities:
raise OwnerAuthorizationError(
"OWNER_IDENTITY_ALLOWLIST_REQUIRED", 503
)
if not scopes:
raise OwnerAuthorizationError(
"OWNER_INTAKE_SCOPE_ALLOWLIST_REQUIRED", 503
)
if any(_is_wildcard_scope(item) for item in scopes):
raise OwnerAuthorizationError(
"WILDCARD_INTAKE_SCOPE_REJECTED", 503
)
if "writer:provision" in scopes:
raise OwnerAuthorizationError(
"WRITER_PROVISION_SCOPE_BOOTSTRAP_REJECTED", 503
)
try:
maximum = int(
os.getenv(
"AIOS_GITOPS_OWNER_AUTHORIZATION_MAX_EXPIRY_SECONDS", "900"
)
)
except ValueError:
raise OwnerAuthorizationError(
"BROKER_MAX_EXPIRY_INVALID", 503
) from None
if maximum < 60 or maximum > 1800:
raise OwnerAuthorizationError("BROKER_MAX_EXPIRY_INVALID", 503)
return cls(
repository_identity=os.getenv(
"AIOS_GITOPS_REPOSITORY_IDENTITY", "aios/aios"
).strip(),
state_root=Path(
os.getenv(
"AIOS_GITOPS_OWNER_AUTHORIZATION_STATE_ROOT",
"/var/lib/aios-gitops/owner-authorization-broker",
)
),
allowed_owner_identities=identities,
allowed_intake_scopes=scopes,
allowed_credential_ids=credential_ids,
max_expiry_seconds=maximum,
)
class TaskWriterProvisioner(Protocol):
def provision(
self, raw: Mapping[str, Any], owner: OwnerContext
) -> dict[str, Any]:
...
def utc_now() -> datetime:
return datetime.now(timezone.utc)
def iso_utc(value: datetime) -> str:
return value.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
def canonical_json_bytes(value: Any) -> bytes:
return (
json.dumps(
value,
ensure_ascii=False,
sort_keys=True,
separators=(",", ":"),
)
+ "\n"
).encode("utf-8")
def sha256_json(value: Any) -> str:
return hashlib.sha256(canonical_json_bytes(value)).hexdigest()
def _csv_set(value: str) -> frozenset[str]:
return frozenset(
item.strip() for item in value.split(",") if item.strip()
)
def _is_wildcard_scope(scope: str) -> bool:
return bool(WILDCARD_SCOPE_RE.fullmatch(scope.strip()))
def _atomic_write_json(path: Path, value: Any, mode: int = 0o600) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
if path.parent.is_symlink() or path.is_symlink():
raise OwnerAuthorizationError("BROKER_STATE_SYMLINK_REJECTED", 500)
fd, temporary = tempfile.mkstemp(prefix=".tmp-", dir=path.parent)
try:
os.fchmod(fd, mode)
with os.fdopen(fd, "wb", closefd=True) as stream:
stream.write(canonical_json_bytes(value))
stream.flush()
os.fsync(stream.fileno())
os.replace(temporary, path)
os.chmod(path, mode)
parent_fd = os.open(
path.parent, os.O_RDONLY | getattr(os, "O_DIRECTORY", 0)
)
try:
os.fsync(parent_fd)
finally:
os.close(parent_fd)
finally:
if os.path.exists(temporary):
os.unlink(temporary)
def _read_json(path: Path) -> dict[str, Any]:
if path.is_symlink() or not path.is_file():
raise OwnerAuthorizationError("BROKER_RECEIPT_INVALID", 500)
if path.stat().st_size <= 0 or path.stat().st_size > 1_048_576:
raise OwnerAuthorizationError("BROKER_RECEIPT_INVALID", 500)
try:
value = json.loads(path.read_text(encoding="utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError):
raise OwnerAuthorizationError(
"BROKER_RECEIPT_JSON_INVALID", 500
) from None
if not isinstance(value, dict):
raise OwnerAuthorizationError("BROKER_RECEIPT_INVALID", 500)
return value
def _prepare_state_root(root: Path) -> None:
if root.is_symlink():
raise OwnerAuthorizationError("BROKER_STATE_SYMLINK_REJECTED", 500)
root.mkdir(parents=True, exist_ok=True, mode=0o700)
os.chmod(root, 0o700)
if root.stat().st_mode & 0o077:
raise OwnerAuthorizationError(
"BROKER_STATE_PERMISSIONS_REJECTED", 500
)
def _event_id(
principal: OwnerPrincipal, request_digest: str
) -> str:
material = (
f"{principal.identity}\0{principal.authenticated_scope}\0"
f"{request_digest}"
).encode("utf-8")
return "owner-auth-" + hashlib.sha256(material).hexdigest()[:32]
def _authorization_reference(event_id: str, request_digest: str) -> str:
return f"owner-auth:{event_id}:{request_digest[:16]}"
def _validate_principal(
principal: OwnerPrincipal,
config: OwnerAuthorizationBrokerConfig,
) -> None:
if principal.role != "owner":
raise OwnerAuthorizationError("OWNER_ROLE_REQUIRED", 403)
if (
not principal.identity
or not IDENTITY_RE.fullmatch(principal.identity)
or principal.identity not in config.allowed_owner_identities
):
raise OwnerAuthorizationError("OWNER_IDENTITY_NOT_ALLOWED", 403)
if (
not principal.authenticated_scope
or not SCOPE_RE.fullmatch(principal.authenticated_scope)
or principal.authenticated_scope
not in config.allowed_intake_scopes
):
raise OwnerAuthorizationError("OWNER_INTAKE_SCOPE_REJECTED", 403)
if principal.authenticated_scope == "writer:provision":
raise OwnerAuthorizationError(
"WRITER_PROVISION_SCOPE_BOOTSTRAP_REJECTED", 403
)
scopes = principal.token_scopes
if any(_is_wildcard_scope(scope) for scope in scopes):
raise OwnerAuthorizationError(
"WILDCARD_OWNER_CREDENTIAL_REJECTED", 403
)
if scopes:
if principal.authenticated_scope not in scopes:
raise OwnerAuthorizationError(
"OWNER_SCOPE_EVIDENCE_MISMATCH", 403
)
elif (
not principal.credential_id
or principal.credential_id not in config.allowed_credential_ids
):
# handler.need(scope) proves authorization but may hide whether the
# credential was wildcard. When scope metadata is absent, require an
# exact non-secret credential ID allowlist as the alternative proof.
raise OwnerAuthorizationError(
"OWNER_CREDENTIAL_SCOPE_NOT_PROVABLE", 403
)
if principal.credential_id:
if not CREDENTIAL_ID_RE.fullmatch(principal.credential_id):
raise OwnerAuthorizationError("OWNER_CREDENTIAL_ID_INVALID", 403)
if (
config.allowed_credential_ids
and principal.credential_id not in config.allowed_credential_ids
):
raise OwnerAuthorizationError(
"OWNER_CREDENTIAL_ID_NOT_ALLOWED", 403
)
def _validate_request(
raw: Mapping[str, Any],
config: OwnerAuthorizationBrokerConfig,
) -> dict[str, Any]:
if set(raw) != BROKER_REQUEST_FIELDS:
raise OwnerAuthorizationError("BROKER_REQUEST_FIELDS_REJECTED")
if raw["action"] != ACTION:
raise OwnerAuthorizationError("BROKER_ACTION_REJECTED")
if raw["repository"] != config.repository_identity:
raise OwnerAuthorizationError(
"REPOSITORY_IDENTITY_MISMATCH", 409
)
if raw["base_ref"] != "refs/heads/main":
raise OwnerAuthorizationError("BASE_REF_REJECTED", 409)
if (
not isinstance(raw["base_oid"], str)
or not OID_RE.fullmatch(raw["base_oid"])
):
raise OwnerAuthorizationError("BASE_OID_REJECTED")
if (
not isinstance(raw["target_branch"], str)
or not BRANCH_RE.fullmatch(raw["target_branch"])
or raw["target_branch"] in {"main", "master"}
):
raise OwnerAuthorizationError("TARGET_BRANCH_REJECTED")
if (
not isinstance(raw["task_id"], str)
or not TASK_RE.fullmatch(raw["task_id"])
):
raise OwnerAuthorizationError("TASK_ID_REJECTED")
if raw["one_time_use"] is not True:
raise OwnerAuthorizationError("ONE_TIME_USE_REQUIRED")
expiry = raw["expires_in_seconds"]
if (
not isinstance(expiry, int)
or isinstance(expiry, bool)
or not 60 <= expiry <= config.max_expiry_seconds
):
raise OwnerAuthorizationError("GRANT_LIFETIME_REJECTED")
for field in (
"production_change_authorized",
"tag_change_authorized",
"baseline_change_authorized",
"main_write_authorized",
):
if raw[field] is not False:
raise OwnerAuthorizationError(
"PRODUCTION_CAPABILITY_REJECTED", 403
)
allowed = raw["allowed_operations"]
denied = raw["denied_operations"]
if (
not isinstance(allowed, list)
or frozenset(allowed) != REQUIRED_ALLOWED_OPERATIONS
):
raise OwnerAuthorizationError("ALLOWED_OPERATIONS_MISMATCH", 403)
if (
not isinstance(denied, list)
or frozenset(denied) != REQUIRED_DENIED_OPERATIONS
):
raise OwnerAuthorizationError("DENIED_OPERATIONS_MISMATCH", 403)
return dict(raw)
def _response_is_safe(value: Any) -> bool:
"""Reject accidental credential material before writing a receipt."""
sensitive_markers = (
"access_token",
"bearer",
"client_secret",
"private_key",
"raw_token",
"secret_value",
)
if isinstance(value, Mapping):
for key, inner in value.items():
lowered = str(key).lower()
if any(marker in lowered for marker in sensitive_markers):
return False
if not _response_is_safe(inner):
return False
elif isinstance(value, list):
return all(_response_is_safe(item) for item in value)
return True
class OwnerAuthorizationBroker:
"""Create an audited Owner event and invoke the governed bridge."""
def __init__(
self,
config: OwnerAuthorizationBrokerConfig | None = None,
provisioner: TaskWriterProvisioner | None = None,
) -> None:
self.config = (
config or OwnerAuthorizationBrokerConfig.from_environment()
)
self.provisioner = provisioner or AuthorizedTaskWriterProvisioner()
def authorize_and_provision(
self,
raw: Mapping[str, Any],
principal: OwnerPrincipal,
) -> dict[str, Any]:
config = self.config
_validate_principal(principal, config)
request = _validate_request(raw, config)
_prepare_state_root(config.state_root)
request_digest = sha256_json(request)
event_id = _event_id(principal, request_digest)
reference = _authorization_reference(event_id, request_digest)
event_dir = config.state_root / "events" / event_id
event_dir.mkdir(parents=True, exist_ok=True, mode=0o700)
os.chmod(event_dir, 0o700)
lock_path = event_dir / "event.lock"
lock_path.touch(mode=0o600, exist_ok=True)
os.chmod(lock_path, 0o600)
with lock_path.open("r+b") as lock:
fcntl.flock(lock.fileno(), fcntl.LOCK_EX)
response_path = event_dir / "response.json"
event_path = event_dir / "event.json"
if response_path.exists():
response = _read_json(response_path)
response["owner_authorization_idempotent"] = True
return response
issued_at = utc_now()
expires_at = issued_at + timedelta(
seconds=int(request["expires_in_seconds"])
)
event: dict[str, Any] = {
"schema": SCHEMA,
"event_id": event_id,
"authorization_reference": reference,
"status": "AUTHORIZED",
"owner": {
"identity": principal.identity,
"role": principal.role,
"authenticated_scope": principal.authenticated_scope,
"credential_id": principal.credential_id or None,
"token_scope_digest_sha256": sha256_json(
sorted(principal.token_scopes)
),
"wildcard_scope_used": False,
},
"request": {
"task_id": request["task_id"],
"repository": request["repository"],
"base_ref": request["base_ref"],
"base_oid": request["base_oid"],
"target_branch": request["target_branch"],
"one_time_use": True,
"production_change_authorized": False,
"tag_change_authorized": False,
"baseline_change_authorized": False,
"main_write_authorized": False,
"allowed_operations": sorted(
REQUIRED_ALLOWED_OPERATIONS
),
"denied_operations": sorted(
REQUIRED_DENIED_OPERATIONS
),
},
"request_digest_sha256": request_digest,
"issued_at": iso_utc(issued_at),
"expires_at": iso_utc(expires_at),
}
event["event_digest_sha256"] = sha256_json(event)
_atomic_write_json(event_path, event)
bridge_request = dict(request)
bridge_request.pop("action")
bridge_request["owner_authorization_reference"] = reference
try:
result = self.provisioner.provision(
bridge_request,
OwnerContext(
identity=principal.identity,
role=principal.role,
authorization_reference=reference,
),
)
if not isinstance(result, dict):
raise OwnerAuthorizationError(
"PROVISIONER_RESPONSE_INVALID", 502
)
if result.get("status") != "READY_FOR_CODE_AUTHORING":
raise OwnerAuthorizationError(
"PROVISIONER_NOT_READY",
409,
details={"status": result.get("status")},
)
if not _response_is_safe(result):
raise OwnerAuthorizationError(
"PROVISIONER_RESPONSE_CONTAINS_SENSITIVE_DATA", 500
)
except ProvisioningError as exc:
event.update(
{
"status": "PROVISIONING_BLOCKED",
"failed_at": iso_utc(utc_now()),
"error": exc.code,
"error_status": exc.status,
}
)
event["event_digest_sha256"] = sha256_json(
{
key: value
for key, value in event.items()
if key != "event_digest_sha256"
}
)
_atomic_write_json(event_path, event)
raise
except OwnerAuthorizationError as exc:
event.update(
{
"status": "PROVISIONING_BLOCKED",
"failed_at": iso_utc(utc_now()),
"error": exc.code,
"error_status": exc.status,
}
)
event["event_digest_sha256"] = sha256_json(
{
key: value
for key, value in event.items()
if key != "event_digest_sha256"
}
)
_atomic_write_json(event_path, event)
raise
except Exception as exc:
event.update(
{
"status": "PROVISIONING_FAILED",
"failed_at": iso_utc(utc_now()),
"error": "OWNER_AUTHORIZATION_BROKER_INTERNAL_ERROR",
}
)
event["event_digest_sha256"] = sha256_json(
{
key: value
for key, value in event.items()
if key != "event_digest_sha256"
}
)
_atomic_write_json(event_path, event)
raise OwnerAuthorizationError(
"OWNER_AUTHORIZATION_BROKER_INTERNAL_ERROR", 500
) from exc
completed_at = utc_now()
event.update(
{
"status": "CONSUMED",
"consumed_at": iso_utc(completed_at),
"approval_plan_id": result.get("approval_plan_id"),
"grant_id": result.get("grant_id"),
"grant_digest_sha256": result.get(
"grant_digest_sha256"
),
"branch_oid": result.get("branch_oid"),
"grant_consumed": result.get("grant_consumed") is True,
}
)
event["event_digest_sha256"] = sha256_json(
{
key: value
for key, value in event.items()
if key != "event_digest_sha256"
}
)
_atomic_write_json(event_path, event)
response = dict(result)
response.update(
{
"owner_authorization_event_id": event_id,
"owner_authorization_reference": reference,
"owner_authorization_digest_sha256": event[
"event_digest_sha256"
],
"owner_authorization_consumed": True,
"owner_authorization_idempotent": False,
}
)
_atomic_write_json(response_path, response)
return response
@@ -0,0 +1,32 @@
"""Read-only release reconciliation and durable follow-up intent."""
from __future__ import annotations
from dataclasses import dataclass
from typing import Any, Mapping
@dataclass(frozen=True)
class ReconciliationReport:
healthy: bool
drift: tuple[str, ...]
actions: tuple[str, ...]
class ReconciliationController:
CHECKS = (
("authority_main", "gitea_main", "main_drift"),
("local_tag_oid", "remote_tag_oid", "tag_drift"),
("remote_peeled_target", "approved_target_oid", "peeled_target_drift"),
("registry_version", "baseline_version", "registry_baseline_drift"),
("approved_artifact_digest", "deployed_artifact_digest", "artifact_drift"),
("declared_production_state", "service_healthy", "service_health_drift"),
)
def reconcile(self, actual: Mapping[str, Any]) -> ReconciliationReport:
drift: list[str] = []
for left, right, label in self.CHECKS:
if actual.get(left) != actual.get(right):
drift.append(label)
actions = () if not drift else ("open_reconciliation_task", "open_reconciliation_pr", "attach_evidence", "notify_owner")
return ReconciliationReport(not drift, tuple(drift), actions)
@@ -0,0 +1,65 @@
"""Durable-friendly recurring task scheduling primitives."""
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from typing import Any
class RecurringTaskError(ValueError):
pass
@dataclass
class RecurringTask:
task_id: str
interval_seconds: int
next_run_at: datetime
run_count: int = 0
max_runs: int | None = None
cancelled: bool = False
def due(self, now: datetime) -> bool:
return not self.cancelled and (self.max_runs is None or self.run_count < self.max_runs) and now >= self.next_run_at
class RecurringTaskScheduler:
def __init__(self) -> None:
self.tasks: dict[str, RecurringTask] = {}
self.successors: set[str] = set()
def schedule(self, task_id: str, interval_seconds: int, *, now: datetime | None = None, max_runs: int | None = None) -> RecurringTask:
if task_id in self.tasks:
raise RecurringTaskError("task already exists")
if interval_seconds <= 0 or (max_runs is not None and max_runs <= 0):
raise RecurringTaskError("invalid recurring task bounds")
now = now or datetime.now(timezone.utc)
task = RecurringTask(task_id, interval_seconds, now, 0, max_runs, False)
self.tasks[task_id] = task
return task
def due(self, *, now: datetime | None = None) -> list[RecurringTask]:
now = now or datetime.now(timezone.utc)
return [task for task in self.tasks.values() if task.due(now)]
def mark_run(self, task_id: str, *, now: datetime | None = None) -> RecurringTask:
task = self.tasks[task_id]
now = now or datetime.now(timezone.utc)
if not task.due(now):
raise RecurringTaskError("task is not due")
task.run_count += 1
task.next_run_at = now + timedelta(seconds=task.interval_seconds)
return task
def cancel(self, task_id: str) -> None:
self.tasks[task_id].cancelled = True
def create_successor_once(self, predecessor: str, successor: str, *, interval_seconds: int, now: datetime | None = None) -> RecurringTask:
key = f"{predecessor}->{successor}"
if key in self.successors:
return self.tasks[successor]
self.successors.add(key)
if successor not in self.tasks:
return self.schedule(successor, interval_seconds, now=now)
return self.tasks[successor]
@@ -0,0 +1,53 @@
"""Release state machine with explicit fail-closed production gates."""
from __future__ import annotations
from enum import Enum
from typing import Mapping
class InvalidTransition(RuntimeError):
pass
class ReleaseState(str, Enum):
SOURCE_READY = "SOURCE_READY"
TESTED = "TESTED"
PR_OPEN = "PR_OPEN"
PR_APPROVED = "PR_APPROVED"
MERGED = "MERGED"
ARTIFACT_VERIFIED = "ARTIFACT_VERIFIED"
RELEASE_APPROVED = "RELEASE_APPROVED"
TAG_VERIFIED = "TAG_VERIFIED"
DEPLOYED = "DEPLOYED"
PRODUCTION_ACCEPTED = "PRODUCTION_ACCEPTED"
TRANSITIONS = {
ReleaseState.SOURCE_READY: (ReleaseState.TESTED,),
ReleaseState.TESTED: (ReleaseState.PR_OPEN,),
ReleaseState.PR_OPEN: (ReleaseState.PR_APPROVED,),
ReleaseState.PR_APPROVED: (ReleaseState.MERGED,),
ReleaseState.MERGED: (ReleaseState.ARTIFACT_VERIFIED,),
ReleaseState.ARTIFACT_VERIFIED: (ReleaseState.RELEASE_APPROVED,),
ReleaseState.RELEASE_APPROVED: (ReleaseState.TAG_VERIFIED,),
ReleaseState.TAG_VERIFIED: (ReleaseState.DEPLOYED,),
ReleaseState.DEPLOYED: (ReleaseState.PRODUCTION_ACCEPTED,),
}
class ReleaseStateMachine:
def __init__(self, state: ReleaseState = ReleaseState.SOURCE_READY) -> None:
self.state = state
def transition(self, target: ReleaseState, evidence: Mapping[str, object] | None = None) -> ReleaseState:
if target not in TRANSITIONS.get(self.state, ()):
raise InvalidTransition(f"{self.state.value}_to_{target.value}_denied")
evidence = evidence or {}
if target is ReleaseState.PRODUCTION_ACCEPTED:
if evidence.get("tag_verified") is not True or evidence.get("remote_tag_readback") is not True or evidence.get("service_health") is not True:
raise InvalidTransition("production_acceptance_evidence_incomplete")
if target is ReleaseState.DEPLOYED and evidence.get("artifact_verified") is not True:
raise InvalidTransition("deployment_artifact_not_verified")
self.state = target
return self.state
@@ -0,0 +1,520 @@
"""Compatibility adapter for the deployed signer and validator contracts.
The bridge keeps authorization and canonical grant construction in the broker,
then translates that internal request into the directory contract consumed by
the deployed root signer and validator. No key material is read by this
module; the signer owns all signing operations.
"""
from __future__ import annotations
import hashlib
import json
import os
import re
import subprocess
from pathlib import Path
from typing import Any, Mapping, Sequence
_SIGNER_FLAGS = (
"--namespace",
"--input",
"--signature-out",
"--key-id-out",
)
class SignerContractError(RuntimeError):
"""Raised when the deployed contract cannot be normalized safely."""
def canonical_json_bytes(value: Any) -> bytes:
return (
json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+ "\n"
).encode("utf-8")
def sha256_json(value: Any) -> str:
return hashlib.sha256(canonical_json_bytes(value)).hexdigest()
def resolve_ref_oid(git_dir: str | Path, ref: str) -> str:
"""Resolve an authoritative commit ref to a full 40-character OID."""
if not isinstance(ref, str) or not re.fullmatch(r"refs/[A-Za-z0-9._/-]+", ref):
raise SignerContractError("authoritative_ref_invalid")
result = subprocess.run(
[
"git",
"--git-dir",
str(git_dir),
"rev-parse",
"--verify",
f"{ref}^{{commit}}",
],
text=True,
capture_output=True,
check=False,
timeout=15,
)
if result.returncode != 0:
raise SignerContractError("authoritative_ref_unavailable")
oid = result.stdout.strip().lower()
if not re.fullmatch(r"[0-9a-f]{40}", oid):
raise SignerContractError("authoritative_ref_invalid_oid")
return oid
def _read(path: Path) -> dict[str, Any]:
if path.is_symlink() or not path.is_file():
raise SignerContractError("contract_input_missing")
try:
value = json.loads(path.read_text(encoding="utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError) as exc:
raise SignerContractError("contract_input_invalid") from exc
if not isinstance(value, dict):
raise SignerContractError("contract_object_required")
return value
def _write(path: Path, value: Any) -> None:
if path.is_symlink():
raise SignerContractError("contract_output_symlink")
path.parent.mkdir(parents=True, exist_ok=True)
temporary = path.with_name(path.name + ".tmp")
temporary.write_bytes(canonical_json_bytes(value))
os.chmod(temporary, 0o600)
os.replace(temporary, path)
os.chmod(path, 0o600)
def deployed_signer_help(stdout: str, stderr: str) -> bool:
text = (stdout or "") + "\n" + (stderr or "")
return all(flag in text for flag in _SIGNER_FLAGS)
def _supports_deployed_signer(command: Sequence[str]) -> bool:
result = subprocess.run(
[*command, "sign", "--help"],
text=True,
capture_output=True,
check=False,
timeout=15,
)
return deployed_signer_help(result.stdout, result.stderr)
def _run(
argv: Sequence[str],
*,
env: Mapping[str, str] | None = None,
timeout: int = 60,
) -> subprocess.CompletedProcess[str]:
result = subprocess.run(
list(argv),
text=True,
capture_output=True,
check=False,
timeout=timeout,
env=dict(env) if env is not None else None,
)
if result.returncode != 0:
raise SignerContractError(
f"deployed_tool_failed:{Path(argv[0]).name}:{result.returncode}"
)
return result
def _sign(
command: Sequence[str],
namespace: str,
input_path: Path,
signature_path: Path,
key_id_path: Path,
) -> None:
if signature_path.exists() or key_id_path.exists():
raise SignerContractError("contract_output_exists")
_run(
[
*command,
"sign",
"--namespace",
namespace,
"--input",
str(input_path),
"--signature-out",
str(signature_path),
"--key-id-out",
str(key_id_path),
],
timeout=60,
)
if not signature_path.is_file() or not key_id_path.is_file():
raise SignerContractError("signer_output_missing")
if signature_path.stat().st_size <= 0 or key_id_path.stat().st_size <= 0:
raise SignerContractError("signer_output_empty")
def _manifest(grant_dir: Path, names: Sequence[str]) -> None:
rows = []
for name in names:
path = grant_dir / name
if not path.is_file() or path.is_symlink():
raise SignerContractError("manifest_input_missing")
rows.append(
f"{hashlib.sha256(path.read_bytes()).hexdigest()} {name}"
)
(grant_dir / "MANIFEST.sha256").write_text(
"\n".join(rows) + "\n",
encoding="utf-8",
)
os.chmod(grant_dir / "MANIFEST.sha256", 0o600)
def _prepare(grant_dir: Path) -> dict[str, Any]:
plan = _read(grant_dir / "approval-plan.json")
request = _read(grant_dir / "grant-request.json")
base = request.get("base")
target = request.get("target")
owner = request.get("owner_authorization")
authorization = request.get("authorization")
if not isinstance(base, dict) or not isinstance(target, dict):
raise SignerContractError("nested_binding_missing")
if not isinstance(owner, dict) or not isinstance(authorization, dict):
raise SignerContractError("authorization_binding_missing")
request_id = "request-" + sha256_json(request)[:32]
credential_request = {
"schema": "aios.credential-request.v1",
"request_id": request_id,
"repository": request["repository"],
"target_branch": target["branch"],
"base_ref": base["ref"],
"base_oid": base["oid"],
"nonce": request["nonce"],
"one_time_use": request["one_time_use"],
"allowed_operations": request["allowed_operations"],
"denied_operations": request["denied_operations"],
"not_before": request["not_before"],
"expires_at": request["expires_at"],
"owner_identity": owner["identity"],
"owner_authorization_reference": owner["reference"],
"production_change_authorized": authorization["production_change_authorized"],
"tag_change_authorized": authorization["tag_change_authorized"],
"baseline_change_authorized": authorization["baseline_change_authorized"],
"main_write_authorized": authorization["main_write_authorized"],
}
request_digest = sha256_json(credential_request)
approval_plan = dict(plan)
approval_plan.update(
{
"repository": request["repository"],
"target_branch": target["branch"],
"base_ref": base["ref"],
"base_oid": base["oid"],
"allowed_operations": request["allowed_operations"],
"denied_operations": request["denied_operations"],
"one_time_use": True,
"max_uses": 1,
"not_before": request["not_before"],
"expires_at": request["expires_at"],
}
)
grant_id = "grant-" + hashlib.sha256(
(request_id + "\0" + request["nonce"]).encode("utf-8")
).hexdigest()[:32]
grant_payload = {
"schema": "aios.grant-payload.v1",
"grant_id": grant_id,
"request_id": request_id,
"request_nonce": request["nonce"],
"credential_request_sha256": request_digest,
"credential_secret_ref": f"task-writer:{grant_id}",
"approval_plan_id": approval_plan["approval_plan_id"],
"repository": request["repository"],
"target_branch": target["branch"],
"base_ref": base["ref"],
"base_oid": base["oid"],
"allowed_operations": request["allowed_operations"],
"denied_operations": request["denied_operations"],
"one_time_use": True,
"max_uses": 1,
"not_before": request["not_before"],
"expires_at": request["expires_at"],
}
grant_digest = sha256_json(grant_payload)
_write(grant_dir / "credential-request.json", credential_request)
(grant_dir / "credential-request.canonical.json").write_bytes(
canonical_json_bytes(credential_request)
)
os.chmod(grant_dir / "credential-request.canonical.json", 0o600)
_write(grant_dir / "approval-plan.json", approval_plan)
(grant_dir / "approval-plan.canonical.json").write_bytes(
canonical_json_bytes(approval_plan)
)
os.chmod(grant_dir / "approval-plan.canonical.json", 0o600)
_write(grant_dir / "grant-payload.json", grant_payload)
(grant_dir / "grant-payload.canonical.json").write_bytes(
canonical_json_bytes(grant_payload)
)
os.chmod(grant_dir / "grant-payload.canonical.json", 0o600)
_write(
grant_dir / "signed-grant.json",
{
"schema": "aios.signed-grant.v1",
"grant_id": grant_id,
"credential_request_sha256": request_digest,
"signed_payload_sha256": grant_digest,
},
)
return {
"request": credential_request,
"approval_plan": approval_plan,
"grant": grant_payload,
"request_digest": request_digest,
"grant_digest": grant_digest,
"grant_id": grant_id,
}
def run_compatible_signer(
signer_command: Sequence[str],
grant_dir: Path,
result_path: Path,
) -> dict[str, str]:
try:
if not _supports_deployed_signer(signer_command):
_run(
[
*signer_command,
"--grant-dir",
str(grant_dir),
"--result-json",
str(result_path),
],
timeout=60,
)
legacy = _read(result_path)
if legacy.get("status") not in {"SIGNED", "PASS", "COMPLETED"}:
raise SignerContractError("legacy_signer_not_ready")
grant_id = legacy.get("grant_id")
digest = legacy.get("grant_digest_sha256")
signer = legacy.get("signing_key_identity") or legacy.get(
"signer_identity"
)
if not all(
isinstance(value, str) and value.strip()
for value in (grant_id, digest, signer)
):
raise SignerContractError("legacy_signer_result_incomplete")
if not legacy.get("signature_present"):
raise SignerContractError("legacy_signature_missing")
return {
"grant_id": grant_id,
"grant_digest_sha256": digest,
"signing_key_identity": signer,
}
prepared = _prepare(grant_dir)
config_path = Path(
os.environ.get("AIOS_GRANT_ROOT_CONFIG", "/etc/aios/grant-root/config.json")
)
config = _read(config_path)
grant_namespace = str(config["grant_namespace"])
approval_namespace = str(config["approval_namespace"])
_sign(
signer_command,
approval_namespace,
grant_dir / "approval-plan.canonical.json",
grant_dir / "approval-plan.sig",
grant_dir / "approval-plan.key-id",
)
_sign(
signer_command,
grant_namespace,
grant_dir / "grant-payload.canonical.json",
grant_dir / "signed-grant.sig",
grant_dir / "signed-grant.key-id",
)
key_identity = (
grant_dir / "signed-grant.key-id"
).read_text(encoding="utf-8").strip()
if not key_identity:
raise SignerContractError("signer_identity_missing")
_manifest(
grant_dir,
(
"approval-plan.json",
"approval-plan.canonical.json",
"approval-plan.sig",
"approval-plan.key-id",
"credential-request.json",
"credential-request.canonical.json",
"grant-payload.json",
"grant-payload.canonical.json",
"signed-grant.json",
"signed-grant.sig",
"signed-grant.key-id",
),
)
result = {
"schema": "aios.compatible-signer-result.v1",
"status": "SIGNED",
"approval_plan_id": prepared["approval_plan"]["approval_plan_id"],
"grant_id": prepared["grant_id"],
"grant_digest_sha256": prepared["grant_digest"],
"signing_key_identity": key_identity,
"signer_identity": key_identity,
"signature_present": True,
}
_write(result_path, result)
return {
"grant_id": prepared["grant_id"],
"grant_digest_sha256": prepared["grant_digest"],
"signing_key_identity": key_identity,
}
except (OSError, KeyError, TypeError, SignerContractError) as exc:
_write(
result_path,
{
"schema": "aios.compatible-signer-result.v1",
"status": "BLOCKED",
"error_code": str(exc),
},
)
raise
def _validator_config(
grant_dir: Path,
request: Mapping[str, Any],
) -> Path:
root_config = Path(
os.environ.get("AIOS_GRANT_ROOT_CONFIG", "/etc/aios/grant-root/config.json")
)
config = _read(root_config)
repo_path = (
os.environ.get("AIOS_GITOPS_VALIDATOR_REPO_PATH")
or os.environ.get("AIOS_GITOPS_WORKSPACE")
or config.get("repo_path")
)
config.update(
{
"repository": request["repository"],
"target_branch": request["target_branch"],
"base_ref": request["base_ref"],
"repo_path": repo_path,
"approval_plan_file": str(grant_dir / "approval-plan.json"),
"approval_plan_canonical_file": str(
grant_dir / "approval-plan.canonical.json"
),
"approval_plan_signature_file": str(grant_dir / "approval-plan.sig"),
"consumed_grants_ledger": str(
grant_dir / "consumed-grants.jsonl"
),
}
)
path = grant_dir.parent / "compat-validator-config.json"
_write(path, config)
return path
def run_compatible_validator(
validator_command: Sequence[str],
grant_dir: Path,
result_path: Path,
) -> Path:
if not (grant_dir / "grant-payload.json").is_file():
try:
_run(
[
*validator_command,
"--grant-dir",
str(grant_dir),
"--result-json",
str(result_path),
],
timeout=60,
)
legacy = _read(result_path)
if legacy.get("status") not in {"PASS", "VALID", "COMPLETED"}:
raise SignerContractError("legacy_validator_not_ready")
return Path(
os.environ.get(
"AIOS_GRANT_ROOT_CONFIG", "/etc/aios/grant-root/config.json"
)
)
except (OSError, KeyError, TypeError, SignerContractError) as exc:
_write(
result_path,
{
"schema": "aios.compatible-validator-result.v1",
"status": "BLOCKED",
"error_code": str(exc),
},
)
raise
request = _read(grant_dir / "credential-request.json")
config_path = _validator_config(grant_dir, request)
env = os.environ.copy()
env["AIOS_GRANT_ROOT_CONFIG"] = str(config_path)
try:
_run(
[
*validator_command,
"--grant-dir",
str(grant_dir),
"--result-json",
str(result_path),
],
env=env,
timeout=60,
)
actual = _read(result_path)
if actual.get("result") != "READY":
raise SignerContractError("validator_not_ready")
checks = {
"schema_valid": True,
"signature_valid": actual.get("signature_verified") is True,
"signer_trusted": actual.get("signature_verified") is True,
"approval_plan_active": actual.get("approval_plan_verified") is True,
"repository_match": True,
"base_ref_match": True,
"base_oid_match": actual.get("remote_base_verified") is True,
"target_branch_match": actual.get("target_branch_absent") is True,
"one_time_use": True,
"allowed_operations_match": True,
"denied_operations_match": True,
}
if not all(checks.values()):
raise SignerContractError("validator_checks_incomplete")
grant = _read(grant_dir / "grant-payload.json")
result = {
"schema": "aios.compatible-validator-result.v1",
"status": "VALID",
"approval_plan_id": grant["approval_plan_id"],
"grant_digest_sha256": sha256_json(grant),
"checks": checks,
"actual_result": {
"result": actual.get("result"),
"signature_verified": actual.get("signature_verified"),
"approval_plan_verified": actual.get("approval_plan_verified"),
"remote_base_verified": actual.get("remote_base_verified"),
"target_branch_absent": actual.get("target_branch_absent"),
"grant_unused": actual.get("grant_unused"),
},
}
_write(result_path, result)
return config_path
except (OSError, KeyError, TypeError, SignerContractError) as exc:
_write(
result_path,
{
"schema": "aios.compatible-validator-result.v1",
"status": "BLOCKED",
"error_code": str(exc),
},
)
raise
+260 -57
View File
@@ -46,6 +46,57 @@ def redact(value: Any) -> str:
return text
class EnvironmentAllowList:
"""Build a subprocess environment without inheriting process secrets."""
SENSITIVE_NAME_RE = re.compile(
r"(?i)(authorization|credential|password|secret|token|api[_-]?key|"
r"askpass|auth[_-]?sock|ld_preload|pythonpath|database_url)"
)
def __init__(self, allowed_names: set[str] | frozenset[str]):
self.allowed_names = frozenset(str(name) for name in allowed_names)
def build(self, values: dict[str, str]) -> dict[str, str]:
result: dict[str, str] = {}
for name, raw in values.items():
if (
name not in self.allowed_names
or self.SENSITIVE_NAME_RE.search(name)
):
raise ValueError(f"environment variable is not allowed: {name}")
value = str(raw)
if redact(value) != value:
raise ValueError(
f"environment value is not allowed: {name}"
)
result[name] = value
return result
SAFE_PROCESS_ENVIRONMENT = EnvironmentAllowList(
{
"PATH",
"LANG",
"LC_ALL",
"TZ",
"HOME",
"GIT_OPTIONAL_LOCKS",
"PYTHONDONTWRITEBYTECODE",
}
).build(
{
"PATH": "/usr/bin:/bin",
"LANG": "C.UTF-8",
"LC_ALL": "C.UTF-8",
"TZ": "UTC",
"HOME": "/var/empty",
"GIT_OPTIONAL_LOCKS": "0",
"PYTHONDONTWRITEBYTECODE": "1",
}
)
def _sanitize(value: Any) -> Any:
if isinstance(value, dict):
return {
@@ -510,11 +561,16 @@ class DeploymentController:
ledger: ApprovalLedger,
policy: SilentReleasePolicy,
store: ReleaseCandidateStore | None = None,
*,
allow_legacy_unsigned_approval: bool = False,
):
self.deployment = deployment
self.ledger = ledger
self.policy = policy
self.store = store
self.allow_legacy_unsigned_approval = (
allow_legacy_unsigned_approval
)
def _save(self, candidate: ReleaseCandidate) -> None:
if self.store:
@@ -525,6 +581,11 @@ class DeploymentController:
candidate: ReleaseCandidate,
approval: ApprovalEnvelope,
) -> dict[str, Any]:
if not self.allow_legacy_unsigned_approval:
raise ApprovalError(
"legacy unsigned runtime approval is disabled; "
"use SignedReleasePlanApproval"
)
if candidate.state != ReleaseState.READY_FOR_APPROVAL:
raise ApprovalError("candidate is not ready for approval")
if not self.policy.requires_owner("production_runtime_switch"):
@@ -675,7 +736,7 @@ class CommandReleaseBuilder:
text=True,
capture_output=True,
timeout=self.config.timeout,
env={**os.environ, "PYTHONDONTWRITEBYTECODE": "1"},
env=dict(SAFE_PROCESS_ENVIRONMENT),
)
except (OSError, subprocess.TimeoutExpired) as exc:
raise RuntimeError("release build unavailable") from exc
@@ -741,12 +802,25 @@ class TagController:
adapter: TagAdapter,
policy: SilentReleasePolicy,
ledger: ApprovalLedger | None,
*,
allow_legacy_unsigned_approval: bool = False,
):
self.adapter = adapter
self.policy = policy
self.ledger = ledger
self.allow_legacy_unsigned_approval = (
allow_legacy_unsigned_approval
)
def _require_legacy_test_mode(self) -> None:
if not self.allow_legacy_unsigned_approval:
raise ApprovalError(
"legacy unsigned tag approval is disabled; "
"use SignedReleasePlanApproval"
)
def approve(self, envelope: TagApprovalEnvelope) -> TagApprovalEnvelope:
self._require_legacy_test_mode()
if not self.policy.requires_owner("production_tag"):
raise ApprovalError("production tag is not owner governed")
if not self.policy.is_forbidden("tag_rewrite"):
@@ -757,6 +831,7 @@ class TagController:
def execute(
self, approved: TagApprovalEnvelope
) -> dict[str, Any]:
self._require_legacy_test_mode()
approved.validate()
existing = self.adapter.lookup(approved.version)
expected = (approved.tag_oid, approved.target_oid)
@@ -793,6 +868,7 @@ class TagController:
def verify(
self, approved: TagApprovalEnvelope
) -> dict[str, Any]:
self._require_legacy_test_mode()
actual = self.adapter.verify(approved.version)
expected = (approved.tag_oid, approved.target_oid)
if actual != expected:
@@ -837,7 +913,7 @@ class CommandReleaseProbe:
text=True,
capture_output=True,
timeout=300,
env={**os.environ, "GIT_OPTIONAL_LOCKS": "0"},
env=dict(SAFE_PROCESS_ENVIRONMENT),
)
except (OSError, subprocess.TimeoutExpired) as exc:
raise RuntimeError("COMMAND_UNAVAILABLE") from exc
@@ -1000,6 +1076,13 @@ class SystemdDeploymentConfig:
health_url: str
command_prefix: tuple[str, ...] = ()
dropin_name: str = "aios-gitops-model-api-production.conf"
service_allowlist: tuple[str, ...] = ("aios-gitops-model-api",)
dropin_allowlist: tuple[str, ...] = (
"aios-gitops-model-api-production.conf",
)
deployment_lock: Path = Path(
"/run/lock/aios-silent-release-deployment.lock"
)
class SystemdDeploymentAdapter:
@@ -1007,39 +1090,134 @@ class SystemdDeploymentAdapter:
def __init__(self, config: SystemdDeploymentConfig):
self.config = config
self._lock_handle = None
self._restart_evidence: dict[str, Any] = {}
if self.config.service_name not in self.config.service_allowlist:
raise ValueError("systemd service is not allowlisted")
if (
self.config.dropin_name not in self.config.dropin_allowlist
or "/" in self.config.dropin_name
or "\\" in self.config.dropin_name
):
raise ValueError("systemd drop-in is not allowlisted")
def _acquire_lock(self) -> None:
if self._lock_handle is not None:
raise RuntimeError("deployment lock is already held")
path = self.config.deployment_lock
path.parent.mkdir(parents=True, exist_ok=True)
handle = path.open("a+", encoding="utf-8")
try:
import fcntl
fcntl.flock(handle.fileno(), fcntl.LOCK_EX)
except ImportError:
pass
self._lock_handle = handle
def _release_lock(self) -> None:
handle = self._lock_handle
self._lock_handle = None
if handle is None:
return
try:
import fcntl
fcntl.flock(handle.fileno(), fcntl.LOCK_UN)
except ImportError:
pass
handle.close()
def _service_identity(self) -> dict[str, Any]:
result = subprocess.run(
[
*self.config.command_prefix,
"systemctl",
"show",
self.config.service_name,
"--property=MainPID",
"--property=ActiveEnterTimestampMonotonic",
"--value",
],
check=True,
capture_output=True,
text=True,
env=dict(SAFE_PROCESS_ENVIRONMENT),
)
values = result.stdout.splitlines()
if len(values) < 2:
raise RuntimeError("systemd process identity unavailable")
try:
pid = int(values[0])
started = int(values[1])
except ValueError as exc:
raise RuntimeError(
"systemd process identity is invalid"
) from exc
return {"main_pid": pid, "started_monotonic": started}
def atomic_switch(
self, candidate: ReleaseCandidate
) -> dict[str, Any]:
target = (
self.config.release_root
/ candidate.source_commit
/ "deployment"
/ self.config.dropin_name
).resolve(strict=True)
link = self.config.cutover_link
if link.exists() and not link.is_symlink():
raise RuntimeError("cutover path is not a managed symlink")
previous = os.readlink(link) if link.is_symlink() else None
link.parent.mkdir(parents=True, exist_ok=True)
temporary = link.with_name(f".{link.name}.{os.getpid()}.tmp")
if temporary.exists() or temporary.is_symlink():
temporary.unlink()
os.symlink(str(target), temporary)
os.replace(temporary, link)
return {
"previous_target": previous,
"new_target": str(target),
"cutover_link": str(link),
}
self._acquire_lock()
try:
release_root = self.config.release_root.resolve(strict=True)
raw_target = (
release_root
/ candidate.source_commit
/ "deployment"
/ self.config.dropin_name
)
if raw_target.is_symlink():
raise RuntimeError("release drop-in must not be a symlink")
target = raw_target.resolve(strict=True)
if target.parent.parent.parent != release_root:
raise RuntimeError("release drop-in escapes release root")
link = self.config.cutover_link
if link.exists() and not link.is_symlink():
raise RuntimeError("cutover path is not a managed symlink")
previous = os.readlink(link) if link.is_symlink() else None
if previous:
previous_path = Path(previous)
if not previous_path.is_absolute():
previous_path = link.parent / previous_path
try:
previous_path.resolve().relative_to(release_root)
except ValueError as exc:
raise RuntimeError(
"existing cutover symlink escapes release root"
) from exc
link.parent.mkdir(parents=True, exist_ok=True)
temporary = link.with_name(f".{link.name}.{os.getpid()}.tmp")
if temporary.exists() or temporary.is_symlink():
temporary.unlink()
os.symlink(str(target), temporary)
os.replace(temporary, link)
return {
"previous_target": previous,
"new_target": str(target),
"cutover_link": str(link),
}
except Exception:
self._release_lock()
raise
def restart(self) -> None:
if self._lock_handle is None:
raise RuntimeError("deployment lock is not held")
before = self._service_identity()
command = [
*self.config.command_prefix,
"systemctl",
"daemon-reload",
]
subprocess.run(command, check=True, capture_output=True, text=True)
subprocess.run(
command,
check=True,
capture_output=True,
text=True,
env=dict(SAFE_PROCESS_ENVIRONMENT),
)
subprocess.run(
[
*self.config.command_prefix,
@@ -1050,7 +1228,18 @@ class SystemdDeploymentAdapter:
check=True,
capture_output=True,
text=True,
env=dict(SAFE_PROCESS_ENVIRONMENT),
)
after = self._service_identity()
if (
after["main_pid"] <= 0
or after["started_monotonic"]
<= before["started_monotonic"]
):
raise RuntimeError(
"systemd restart did not produce a new start identity"
)
self._restart_evidence = {"before": before, "after": after}
def health_check(
self, candidate: ReleaseCandidate
@@ -1062,44 +1251,58 @@ class SystemdDeploymentAdapter:
payload = json.loads(response.read().decode("utf-8"))
except Exception as exc:
raise RuntimeError("runtime health check unavailable") from exc
return {
"passed": response.status == 200,
"runtime_commit": payload.get("runtime_release_commit"),
"writes_enabled": payload.get("writes_enabled"),
"authoritative_head_oid": payload.get(
"authoritative_head_oid"
),
"repository_identity": payload.get("repository_identity"),
}
try:
return {
"passed": response.status == 200,
"runtime_commit": payload.get("runtime_release_commit"),
"writes_enabled": payload.get("writes_enabled"),
"authoritative_head_oid": payload.get(
"authoritative_head_oid"
),
"repository_identity": payload.get("repository_identity"),
"systemd_restart": _sanitize(self._restart_evidence),
}
finally:
self._release_lock()
def rollback(
self, switch_evidence: dict[str, Any]
) -> dict[str, Any]:
link = self.config.cutover_link
previous = switch_evidence.get("previous_target")
temporary = link.with_name(f".{link.name}.{os.getpid()}.rollback")
if previous:
if temporary.exists() or temporary.is_symlink():
temporary.unlink()
os.symlink(str(previous), temporary)
os.replace(temporary, link)
elif link.is_symlink():
link.unlink()
self.restart()
try:
with urllib.request.urlopen(
self.config.health_url, timeout=15
) as response:
payload = json.loads(response.read().decode("utf-8"))
except Exception as exc:
raise RuntimeError(
"rollback health check unavailable"
) from exc
return {
"passed": response.status == 200,
"writes_enabled": payload.get("writes_enabled"),
"runtime_commit": payload.get("runtime_release_commit"),
}
if self._lock_handle is None:
self._acquire_lock()
link = self.config.cutover_link
previous = switch_evidence.get("previous_target")
temporary = link.with_name(
f".{link.name}.{os.getpid()}.rollback"
)
if previous:
if temporary.exists() or temporary.is_symlink():
temporary.unlink()
os.symlink(str(previous), temporary)
os.replace(temporary, link)
elif link.is_symlink():
link.unlink()
self.restart()
try:
with urllib.request.urlopen(
self.config.health_url, timeout=15
) as response:
payload = json.loads(
response.read().decode("utf-8")
)
except Exception as exc:
raise RuntimeError(
"rollback health check unavailable"
) from exc
return {
"passed": response.status == 200,
"writes_enabled": payload.get("writes_enabled"),
"runtime_commit": payload.get("runtime_release_commit"),
"systemd_restart": _sanitize(self._restart_evidence),
}
finally:
self._release_lock()
def _sha256(path: Path) -> str:
@@ -0,0 +1,564 @@
#!/usr/bin/env python3
"""Execution-readiness adapters and the fail-closed Shadow controller service."""
from __future__ import annotations
import argparse
import base64
import dataclasses
import hashlib
import json
import os
import subprocess
import threading
import time
from http import HTTPStatus
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from pathlib import Path
from typing import Any, Callable, Mapping, Protocol
from urllib.parse import parse_qs, urlparse
from cryptography.hazmat.primitives.asymmetric.ed25519 import (
Ed25519PrivateKey,
)
from silent_release_controller import _sanitize, utc_now
from silent_release_hardening import (
BuilderAttestation,
CapabilityAuthorizationError,
ReleaseRecoveryWorker,
SignedApprovalVerifier,
SignedReleasePlanApproval,
SilentReleaseActionService,
TrustedCandidateBundle,
_canonical_json,
_durable_json,
_safe_child,
)
class BuilderSigner(Protocol):
@property
def key_id(self) -> str:
...
def sign(self, payload: bytes) -> bytes:
...
class LocalTestBuilderSigner:
"""Ephemeral in-memory signer. It is forbidden for production execution."""
test_only = True
def __init__(self, key_id: str, private_key: Ed25519PrivateKey):
self._key_id = str(key_id)
self._private_key = private_key
@property
def key_id(self) -> str:
return self._key_id
@classmethod
def generate(
cls, key_id: str = "shadow-builder-test-key"
) -> "LocalTestBuilderSigner":
return cls(key_id, Ed25519PrivateKey.generate())
@property
def public_key(self):
return self._private_key.public_key()
def sign(self, payload: bytes) -> bytes:
return self._private_key.sign(payload)
class ExternalKmsBuilderSigner:
"""Signer facade whose private operation remains in KMS/HSM."""
def __init__(
self,
key_id: str,
credential_reference: str,
sign_callback: Callable[[str, bytes], bytes],
):
if not str(credential_reference).startswith(
("kms://", "hsm://", "secret-manager://")
):
raise ValueError("production builder requires a KMS/HSM reference")
self._key_id = str(key_id)
self.credential_reference = str(credential_reference)
self._sign_callback = sign_callback
@property
def key_id(self) -> str:
return self._key_id
def sign(self, payload: bytes) -> bytes:
return bytes(self._sign_callback(self.credential_reference, payload))
class SignedTrustedBuilderAdapter:
"""Signs evidence produced by a trusted collector, never caller evidence."""
def __init__(
self,
collector: Callable[[Mapping[str, Any]], TrustedCandidateBundle],
signer: BuilderSigner,
):
self.collector = collector
self.signer = signer
def build(self, intent: Mapping[str, Any]) -> TrustedCandidateBundle:
forbidden = {
"artifact_sha256",
"canary",
"checks",
"manifest_sha256",
"tests",
"tree_oid",
}
if forbidden.intersection(intent):
raise CapabilityAuthorizationError(
"trusted builder rejects caller-supplied evidence"
)
unsigned = self.collector(dict(intent))
if unsigned.attestation is not None:
raise CapabilityAuthorizationError(
"collector cannot self-attest its evidence"
)
candidate = unsigned.candidate
test_sha = hashlib.sha256(
_canonical_json(_sanitize(candidate.tests))
).hexdigest()
canary_sha = hashlib.sha256(
_canonical_json(_sanitize(candidate.canary))
).hexdigest()
attestation_id = "builder-attestation-" + hashlib.sha256(
(
candidate.candidate_id
+ "\0"
+ candidate.source_commit
+ "\0"
+ candidate.artifact_sha256
).encode("utf-8")
).hexdigest()[:24]
attestation = BuilderAttestation(
builder_attestation_id=attestation_id,
builder_key_id=self.signer.key_id,
source_commit=candidate.source_commit,
tree_oid=candidate.tree_oid,
artifact_sha256=candidate.artifact_sha256,
manifest_sha256=candidate.manifest_sha256,
package_closure_sha256=(
unsigned.release_plan.deployment.package_closure_sha256
),
test_evidence_sha256=test_sha,
canary_evidence_sha256=canary_sha,
issued_at=utc_now(),
signature="",
)
attestation = dataclasses.replace(
attestation,
signature=base64.b64encode(
self.signer.sign(attestation.canonical_bytes())
).decode("ascii"),
)
return dataclasses.replace(unsigned, attestation=attestation)
class ApprovalBackendAdapter(Protocol):
@property
def available(self) -> bool:
...
def verify(
self,
approval: SignedReleasePlanApproval,
*,
candidate,
release_plan,
policy_sha256: str,
) -> SignedReleasePlanApproval:
...
class LocalTestApprovalSigner:
"""Test-only approval signer used solely for Shadow acceptance."""
test_only = True
def __init__(self, key_id: str, private_key: Ed25519PrivateKey):
self.key_id = str(key_id)
self.private_key = private_key
@classmethod
def generate(
cls, key_id: str = "shadow-owner-test-key"
) -> "LocalTestApprovalSigner":
return cls(key_id, Ed25519PrivateKey.generate())
def sign(
self, approval: SignedReleasePlanApproval
) -> SignedReleasePlanApproval:
unsigned = dataclasses.replace(
approval, key_id=self.key_id, signature=""
)
return dataclasses.replace(
unsigned,
signature=base64.b64encode(
self.private_key.sign(unsigned.canonical_bytes())
).decode("ascii"),
)
class ProductionApprovalBackendAdapter:
"""Verification-only production adapter configured by a secret/KMS ref."""
def __init__(
self,
verifier: SignedApprovalVerifier,
credential_reference: str,
):
if not str(credential_reference).startswith(
("kms://", "hsm://", "secret-manager://")
):
raise ValueError(
"production approval backend requires a KMS/secret reference"
)
self.verifier = verifier
self.credential_reference = str(credential_reference)
@property
def available(self) -> bool:
return True
def verify(
self,
approval: SignedReleasePlanApproval,
*,
candidate,
release_plan,
policy_sha256: str,
) -> SignedReleasePlanApproval:
return self.verifier.verify_release_plan(
approval, candidate, release_plan, policy_sha256
)
class UnavailableProductionApprovalBackend:
credential_reference = None
@property
def available(self) -> bool:
return False
def verify(self, *args, **kwargs):
raise CapabilityAuthorizationError(
"production approval backend is not configured"
)
class LeaderLock:
def __init__(self, path: Path):
self.path = Path(path)
self.handle = None
def __enter__(self):
self.path.parent.mkdir(parents=True, exist_ok=True)
os.chmod(self.path.parent, 0o700)
self.handle = self.path.open("a+", encoding="utf-8")
os.chmod(self.path, 0o600)
try:
import fcntl
fcntl.flock(
self.handle.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB
)
except BlockingIOError as exc:
self.handle.close()
self.handle = None
raise RuntimeError("silent release leader is already active") from exc
return self
def __exit__(self, exc_type, exc, tb):
if self.handle is not None:
try:
import fcntl
fcntl.flock(self.handle.fileno(), fcntl.LOCK_UN)
finally:
self.handle.close()
self.handle = None
class PersistentSilentReleaseController:
"""Idempotent main-event controller with durable startup/recovery reports."""
def __init__(
self,
root: Path,
service: SilentReleaseActionService,
*,
main_probe: Callable[[], str],
recovery_worker: ReleaseRecoveryWorker | None = None,
):
self.root = Path(root)
self.service = service
self.main_probe = main_probe
self.recovery_worker = recovery_worker
self.leader_path = self.root / "controller.leader.lock"
self.status_path = self.root / "controller-status.json"
def startup_verify(self) -> dict[str, Any]:
checked = 0
failed = 0
for journal in sorted(self.service.sagas.root.glob("*.events.jsonl")):
candidate_id = journal.name[: -len(".events.jsonl")]
saga = self.service.sagas.load(candidate_id)
checked += 1
if not saga.journal_integrity:
failed += 1
status = {
"checked_sagas": checked,
"execution_enabled": self.service.execution_enabled,
"failed_closed": failed,
"shadow_mode": not self.service.execution_enabled,
"side_effects_executed": False,
"status": "READY" if failed == 0 else "FAIL_CLOSED",
}
_durable_json(self.status_path, status)
return status
def run_once(self) -> dict[str, Any]:
with LeaderLock(self.leader_path):
startup = self.startup_verify()
if startup["status"] != "READY":
result = {
"execution_enabled": self.service.execution_enabled,
"intent": None,
"notification": None,
"recovery": None,
"shadow_mode": not self.service.execution_enabled,
"side_effects_executed": False,
"startup": startup,
"status": "FAIL_CLOSED",
}
_durable_json(self.status_path, result)
return _sanitize(result)
source_commit = str(self.main_probe())
event_id = f"main-{source_commit}"
intent = self.service.create_release_intent(
{"event_id": event_id, "source_commit": source_commit}
)
notification = None
if intent.get("candidate_id"):
notification = self.service.ensure_owner_notification(
str(intent["candidate_id"])
)
recovery = (
self.recovery_worker.run_once()
if self.recovery_worker is not None
else self.service.get_recovery_status()
)
result = {
"execution_enabled": self.service.execution_enabled,
"intent": intent,
"notification": notification,
"recovery": recovery,
"shadow_mode": not self.service.execution_enabled,
"side_effects_executed": False,
"startup": startup,
}
_durable_json(self.status_path, result)
return _sanitize(result)
class ShadowActionHandler(BaseHTTPRequestHandler):
service: SilentReleaseActionService
controller: PersistentSilentReleaseController
def log_message(self, fmt, *args):
return
def _json(self, payload: Mapping[str, Any], status=HTTPStatus.OK):
body = json.dumps(_sanitize(dict(payload)), sort_keys=True).encode()
self.send_response(int(status))
self.send_header("Content-Type", "application/json")
self.send_header("Content-Length", str(len(body)))
self.end_headers()
self.wfile.write(body)
def _body(self) -> dict[str, Any]:
size = int(self.headers.get("Content-Length", "0"))
if size > 1024 * 1024:
raise ValueError("request body is too large")
return json.loads(self.rfile.read(size) or b"{}")
def do_GET(self):
parsed = urlparse(self.path)
query = parse_qs(parsed.query)
candidate_id = str((query.get("candidate_id") or [""])[0])
try:
if parsed.path == "/healthz":
return self._json(
{
"execution_enabled": False,
"shadow_mode": True,
"side_effects_executed": False,
"status": "ok",
}
)
routes = {
"/getReleaseCandidate": self.service.get_release_candidate,
"/getReleasePlan": self.service.get_release_plan,
"/getReleaseAcceptance": self.service.get_acceptance,
"/getReleaseSaga": self.service.get_release_saga,
"/getCapabilityStatus": self.service.get_capability_status,
}
if parsed.path in routes:
return self._json(routes[parsed.path](candidate_id))
if parsed.path == "/getRecoveryStatus":
return self._json(self.service.get_recovery_status())
self._json({"error": "not_found"}, HTTPStatus.NOT_FOUND)
except (ValueError, FileNotFoundError) as exc:
self._json(
{"error": type(exc).__name__}, HTTPStatus.BAD_REQUEST
)
def do_POST(self):
parsed = urlparse(self.path)
try:
body = self._body()
if parsed.path == "/createReleaseIntent":
return self._json(self.service.create_release_intent(body))
if parsed.path in {
"/submitSignedApproval",
"/submitOwnerApproval",
}:
return self._json(self.service.submit_owner_approval(body))
if parsed.path == "/startDeployment":
return self._json(
self.service.start_deployment(
str(body.get("candidate_id") or "")
)
)
self._json({"error": "not_found"}, HTTPStatus.NOT_FOUND)
except (ValueError, FileNotFoundError, CapabilityAuthorizationError) as exc:
self._json(
{"error": type(exc).__name__}, HTTPStatus.BAD_REQUEST
)
def git_main_probe(repository: Path) -> Callable[[], str]:
repository = Path(repository)
def probe() -> str:
completed = subprocess.run(
[
"git",
f"--git-dir={repository}",
"rev-parse",
"refs/heads/main",
],
check=True,
capture_output=True,
text=True,
timeout=15,
env={"PATH": os.environ.get("PATH", "")},
)
return completed.stdout.strip()
return probe
def serve_shadow(
*,
state_root: Path,
repository: Path,
host: str,
port: int,
poll_seconds: float,
) -> None:
service = SilentReleaseActionService.shadow(state_root)
controller = PersistentSilentReleaseController(
state_root,
service,
main_probe=git_main_probe(repository),
)
controller.run_once()
stop = threading.Event()
def controller_loop() -> None:
while not stop.wait(poll_seconds):
try:
controller.run_once()
except Exception as exc: # fail closed and keep the API available
_durable_json(
controller.status_path,
{
"error": type(exc).__name__,
"execution_enabled": False,
"shadow_mode": True,
"side_effects_executed": False,
"status": "FAIL_CLOSED",
},
)
handler = type(
"ConfiguredShadowActionHandler",
(ShadowActionHandler,),
{"service": service, "controller": controller},
)
server = ThreadingHTTPServer((host, port), handler)
api_thread = threading.Thread(
target=server.serve_forever,
name="silent-release-shadow-api",
daemon=False,
)
controller_thread = threading.Thread(
target=controller_loop,
name="silent-release-shadow-controller",
daemon=True,
)
api_thread.start()
controller_thread.start()
try:
while api_thread.is_alive():
api_thread.join(timeout=1)
finally:
stop.set()
server.shutdown()
server.server_close()
controller_thread.join(timeout=max(1.0, poll_seconds + 1.0))
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--state-root", required=True, type=Path)
parser.add_argument("--repository", required=True, type=Path)
parser.add_argument("--host", default="127.0.0.1")
parser.add_argument("--port", default=18891, type=int)
parser.add_argument("--poll-seconds", default=30.0, type=float)
arguments = parser.parse_args()
if arguments.poll_seconds <= 0:
raise SystemExit("poll seconds must be positive")
if os.environ.get("AIOS_GITOPS_WRITES_ENABLED", "false").lower() != "false":
raise SystemExit("GitOps writes must remain disabled")
if (
os.environ.get(
"AIOS_SILENT_RELEASE_EXECUTION_ENABLED", "false"
).lower()
!= "false"
):
raise SystemExit("Silent Release execution must remain disabled")
serve_shadow(
state_root=arguments.state_root,
repository=arguments.repository,
host=arguments.host,
port=arguments.port,
poll_seconds=arguments.poll_seconds,
)
if __name__ == "__main__":
main()
File diff suppressed because it is too large Load Diff
+760
View File
@@ -0,0 +1,760 @@
"""Task-scoped identity overlay for bootstrap writer provisioning.
This module resolves bearer-registry identity ambiguity without modifying the
global token registry. A root-governed one-shot issuer creates a random,
short-lived credential bound to exactly one repository, base OID, task and
target branch. The HTTP overlay authenticates only that credential and invokes
the existing OwnerAuthorizationBroker.
The overlay never grants main, tag, baseline, production or repository-admin
capabilities. The bearer secret is stored only in a root-owned runtime path and
is removed after success, failure, revocation or expiry.
"""
from __future__ import annotations
import fcntl
import hashlib
import json
import os
import re
import secrets
import tempfile
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Mapping, Sequence
from authorized_task_writer_provisioning import (
BRANCH_RE,
ProvisioningError,
OID_RE,
REQUIRED_ALLOWED_OPERATIONS,
REQUIRED_DENIED_OPERATIONS,
TASK_RE,
)
from owner_authorization_broker import (
ACTION,
BROKER_REQUEST_FIELDS,
OwnerAuthorizationBroker,
OwnerAuthorizationBrokerConfig,
OwnerAuthorizationError,
OwnerPrincipal,
TaskWriterProvisioner,
canonical_json_bytes,
iso_utc,
sha256_json,
utc_now,
)
OVERLAY_SCHEMA = "aios.task-identity-overlay.v1"
OVERLAY_SCOPE = "owner:task-authorize"
OVERLAY_ID_RE = re.compile(r"^overlay-[0-9a-f]{32}$")
OWNER_IDENTITY_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._@:+/-]{1,254}$")
AUTH_ASSERTION_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:/@+-]{3,255}$")
TERMINAL_STATES = frozenset({"CONSUMED", "FAILED", "REVOKED", "EXPIRED"})
MAX_BODY_BYTES = 1_048_576
class IdentityOverlayError(RuntimeError):
"""Stable, fail-closed error for the overlay control plane."""
def __init__(
self,
code: str,
status: int = 400,
*,
details: Mapping[str, Any] | None = None,
) -> None:
super().__init__(code)
self.code = code
self.status = status
self.details = dict(details or {})
@dataclass(frozen=True)
class TaskBinding:
task_id: str
repository: str
base_ref: str
base_oid: str
target_branch: str
def as_dict(self) -> dict[str, str]:
return {
"task_id": self.task_id,
"repository": self.repository,
"base_ref": self.base_ref,
"base_oid": self.base_oid,
"target_branch": self.target_branch,
}
@dataclass(frozen=True)
class IdentityOverlayConfig:
state_root: Path
secret_root: Path
binding: TaskBinding
owner_identity: str
max_ttl_seconds: int = 900
require_root_issue: bool = True
collision_secret_refs: tuple[Path, ...] = ()
broker_state_root: Path | None = None
@classmethod
def from_environment(cls) -> "IdentityOverlayConfig":
try:
maximum = int(
os.getenv("AIOS_GITOPS_IDENTITY_OVERLAY_MAX_TTL_SECONDS", "900")
)
except ValueError:
raise IdentityOverlayError("OVERLAY_MAX_TTL_INVALID", 503) from None
if not 60 <= maximum <= 900:
raise IdentityOverlayError("OVERLAY_MAX_TTL_INVALID", 503)
refs = tuple(
Path(item.strip())
for item in os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_COLLISION_SECRET_REFS", ""
).split(",")
if item.strip()
)
owner = os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_OWNER_IDENTITY", ""
).strip()
config = cls(
state_root=Path(
os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_STATE_ROOT",
"/var/lib/aios-gitops/task-identity-overlays",
)
),
secret_root=Path(
os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_SECRET_ROOT",
"/run/aios-gitops/task-identity-overlays",
)
),
broker_state_root=Path(
os.getenv(
"AIOS_GITOPS_OWNER_AUTHORIZATION_STATE_ROOT",
"/var/lib/aios-gitops/owner-authorization-broker",
)
),
binding=TaskBinding(
task_id=os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_TASK_ID", ""
).strip(),
repository=os.getenv(
"AIOS_GITOPS_REPOSITORY_IDENTITY", "aios/aios"
).strip(),
base_ref=os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_BASE_REF",
"refs/heads/main",
).strip(),
base_oid=os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_BASE_OID", ""
).strip(),
target_branch=os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_TARGET_BRANCH", ""
).strip(),
),
owner_identity=owner,
max_ttl_seconds=maximum,
require_root_issue=os.getenv(
"AIOS_GITOPS_IDENTITY_OVERLAY_REQUIRE_ROOT", "true"
).strip().lower()
not in {"0", "false", "no", "off"},
collision_secret_refs=refs,
)
_validate_config(config)
return config
@dataclass(frozen=True)
class OverlaySession:
overlay_id: str
metadata_path: Path
secret_ref: Path
owner_identity: str
request_digest_sha256: str
binding: TaskBinding
def _validate_config(config: IdentityOverlayConfig) -> None:
binding = config.binding
if not TASK_RE.fullmatch(binding.task_id):
raise IdentityOverlayError("OVERLAY_TASK_ID_INVALID", 503)
if binding.repository != "aios/aios":
raise IdentityOverlayError("OVERLAY_REPOSITORY_REJECTED", 503)
if binding.base_ref != "refs/heads/main":
raise IdentityOverlayError("OVERLAY_BASE_REF_REJECTED", 503)
if not OID_RE.fullmatch(binding.base_oid):
raise IdentityOverlayError("OVERLAY_BASE_OID_INVALID", 503)
if (
not BRANCH_RE.fullmatch(binding.target_branch)
or binding.target_branch in {"main", "master"}
):
raise IdentityOverlayError("OVERLAY_TARGET_BRANCH_INVALID", 503)
if not OWNER_IDENTITY_RE.fullmatch(config.owner_identity):
raise IdentityOverlayError("OVERLAY_OWNER_IDENTITY_INVALID", 503)
if not 60 <= config.max_ttl_seconds <= 900:
raise IdentityOverlayError("OVERLAY_MAX_TTL_INVALID", 503)
if config.state_root == config.secret_root:
raise IdentityOverlayError("OVERLAY_STATE_SECRET_ROOT_COLLISION", 503)
def _prepare_directory(path: Path, mode: int = 0o700) -> None:
if path.is_symlink():
raise IdentityOverlayError("OVERLAY_PATH_SYMLINK_REJECTED", 500)
path.mkdir(parents=True, exist_ok=True, mode=mode)
os.chmod(path, mode)
if path.stat().st_mode & 0o077:
raise IdentityOverlayError("OVERLAY_DIRECTORY_PERMISSIONS_REJECTED", 500)
def _atomic_write(path: Path, data: bytes, mode: int) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
if path.parent.is_symlink() or path.is_symlink():
raise IdentityOverlayError("OVERLAY_PATH_SYMLINK_REJECTED", 500)
fd, temporary = tempfile.mkstemp(prefix=".tmp-", dir=path.parent)
try:
os.fchmod(fd, mode)
with os.fdopen(fd, "wb", closefd=True) as stream:
stream.write(data)
stream.flush()
os.fsync(stream.fileno())
os.replace(temporary, path)
os.chmod(path, mode)
parent_fd = os.open(
path.parent, os.O_RDONLY | getattr(os, "O_DIRECTORY", 0)
)
try:
os.fsync(parent_fd)
finally:
os.close(parent_fd)
finally:
if os.path.exists(temporary):
os.unlink(temporary)
def _atomic_write_json(path: Path, value: Any, mode: int = 0o600) -> None:
_atomic_write(path, canonical_json_bytes(value), mode)
def _read_json(path: Path) -> dict[str, Any]:
if path.is_symlink() or not path.is_file():
raise IdentityOverlayError("OVERLAY_METADATA_INVALID", 500)
if path.stat().st_size <= 0 or path.stat().st_size > MAX_BODY_BYTES:
raise IdentityOverlayError("OVERLAY_METADATA_INVALID", 500)
try:
value = json.loads(path.read_text(encoding="utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError):
raise IdentityOverlayError("OVERLAY_METADATA_INVALID", 500) from None
if not isinstance(value, dict):
raise IdentityOverlayError("OVERLAY_METADATA_INVALID", 500)
return value
def _parse_utc(value: str) -> datetime:
try:
parsed = datetime.fromisoformat(value.replace("Z", "+00:00"))
except (TypeError, ValueError):
raise IdentityOverlayError("OVERLAY_TIMESTAMP_INVALID", 500) from None
if parsed.tzinfo is None:
raise IdentityOverlayError("OVERLAY_TIMESTAMP_INVALID", 500)
return parsed.astimezone(timezone.utc)
def _token_hash(token: bytes) -> str:
return hashlib.sha256(token.strip()).hexdigest()
def _safe_unlink(path: Path) -> None:
try:
if path.is_symlink():
raise IdentityOverlayError("OVERLAY_SECRET_SYMLINK_REJECTED", 500)
path.unlink(missing_ok=True)
except OSError as exc:
raise IdentityOverlayError("OVERLAY_SECRET_REMOVAL_FAILED", 500) from exc
def _validate_request_binding(
request: Mapping[str, Any], binding: TaskBinding, max_ttl: int
) -> dict[str, Any]:
if set(request) != BROKER_REQUEST_FIELDS:
raise IdentityOverlayError("OVERLAY_REQUEST_FIELDS_REJECTED")
if request.get("action") != ACTION:
raise IdentityOverlayError("OVERLAY_ACTION_REJECTED")
expected = binding.as_dict()
for key, value in expected.items():
if request.get(key) != value:
raise IdentityOverlayError(
"OVERLAY_BINDING_MISMATCH", 409, details={"field": key}
)
if request.get("one_time_use") is not True:
raise IdentityOverlayError("OVERLAY_ONE_TIME_USE_REQUIRED")
expiry = request.get("expires_in_seconds")
if (
not isinstance(expiry, int)
or isinstance(expiry, bool)
or not 60 <= expiry <= max_ttl
):
raise IdentityOverlayError("OVERLAY_TTL_REJECTED")
for field in (
"production_change_authorized",
"tag_change_authorized",
"baseline_change_authorized",
"main_write_authorized",
):
if request.get(field) is not False:
raise IdentityOverlayError("OVERLAY_PRODUCTION_CAPABILITY_REJECTED", 403)
if (
not isinstance(request.get("allowed_operations"), list)
or frozenset(request["allowed_operations"])
!= REQUIRED_ALLOWED_OPERATIONS
):
raise IdentityOverlayError("OVERLAY_ALLOWED_OPERATIONS_MISMATCH", 403)
if (
not isinstance(request.get("denied_operations"), list)
or frozenset(request["denied_operations"])
!= REQUIRED_DENIED_OPERATIONS
):
raise IdentityOverlayError("OVERLAY_DENIED_OPERATIONS_MISMATCH", 403)
return dict(request)
def _metadata_digest(metadata: Mapping[str, Any]) -> str:
unsigned = {
key: value
for key, value in metadata.items()
if key != "metadata_digest_sha256"
}
return sha256_json(unsigned)
def _verify_metadata(metadata: Mapping[str, Any]) -> None:
if metadata.get("schema") != OVERLAY_SCHEMA:
raise IdentityOverlayError("OVERLAY_METADATA_SCHEMA_INVALID", 500)
overlay_id = str(metadata.get("overlay_id") or "")
if not OVERLAY_ID_RE.fullmatch(overlay_id):
raise IdentityOverlayError("OVERLAY_METADATA_ID_INVALID", 500)
expected = _metadata_digest(metadata)
if metadata.get("metadata_digest_sha256") != expected:
raise IdentityOverlayError("OVERLAY_METADATA_DIGEST_MISMATCH", 500)
class TaskIdentityOverlayStore:
"""Issue, authenticate, consume and revoke one task-bound credential."""
def __init__(self, config: IdentityOverlayConfig) -> None:
_validate_config(config)
self.config = config
@property
def lock_path(self) -> Path:
return self.config.state_root / "overlay.lock"
def _lock(self):
_prepare_directory(self.config.state_root)
_prepare_directory(self.config.secret_root)
self.lock_path.touch(mode=0o600, exist_ok=True)
os.chmod(self.lock_path, 0o600)
stream = self.lock_path.open("r+b")
fcntl.flock(stream.fileno(), fcntl.LOCK_EX)
return stream
def _metadata_paths(self) -> list[Path]:
return sorted(self.config.state_root.glob("overlay-*/metadata.json"))
def _active_for_task(self) -> tuple[Path, dict[str, Any]] | None:
now = utc_now()
for path in self._metadata_paths():
metadata = _read_json(path)
_verify_metadata(metadata)
if metadata.get("task_id") != self.config.binding.task_id:
continue
state = metadata.get("status")
if state in TERMINAL_STATES:
continue
if _parse_utc(str(metadata["expires_at"])) <= now:
metadata["status"] = "EXPIRED"
metadata["expired_at"] = iso_utc(now)
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
_atomic_write_json(path, metadata)
_safe_unlink(Path(str(metadata["secret_ref"])))
continue
return path, metadata
return None
def _assert_hash_unique(self, digest: str) -> None:
matches = 0
for path in self._metadata_paths():
metadata = _read_json(path)
_verify_metadata(metadata)
if metadata.get("token_sha256") == digest:
matches += 1
for secret_ref in self.config.collision_secret_refs:
if not secret_ref.exists():
continue
if secret_ref.is_symlink() or not secret_ref.is_file():
raise IdentityOverlayError(
"OVERLAY_COLLISION_SECRET_REF_INVALID", 500
)
if secret_ref.stat().st_size > 65_536:
raise IdentityOverlayError(
"OVERLAY_COLLISION_SECRET_REF_INVALID", 500
)
if _token_hash(secret_ref.read_bytes()) == digest:
matches += 1
if matches:
raise IdentityOverlayError("OVERLAY_SECRET_IDENTITY_COLLISION", 409)
def issue(
self,
*,
ttl_seconds: int,
authorization_assertion_ref: str,
operator_identity: str,
) -> dict[str, Any]:
if self.config.require_root_issue and os.geteuid() != 0:
raise IdentityOverlayError("OVERLAY_ROOT_ISSUER_REQUIRED", 403)
if (
not isinstance(ttl_seconds, int)
or isinstance(ttl_seconds, bool)
or not 60 <= ttl_seconds <= self.config.max_ttl_seconds
):
raise IdentityOverlayError("OVERLAY_TTL_REJECTED")
if not AUTH_ASSERTION_RE.fullmatch(authorization_assertion_ref):
raise IdentityOverlayError("OVERLAY_AUTH_ASSERTION_REF_INVALID")
if not OWNER_IDENTITY_RE.fullmatch(operator_identity):
raise IdentityOverlayError("OVERLAY_OPERATOR_IDENTITY_INVALID")
with self._lock():
active = self._active_for_task()
if active:
_, metadata = active
expected = self.config.binding.as_dict()
actual = {
key: metadata.get(key)
for key in expected
}
if actual != expected:
raise IdentityOverlayError(
"OVERLAY_ACTIVE_BINDING_CONFLICT", 409
)
return {
"status": "OVERLAY_ACTIVE",
"idempotent": True,
"overlay_id": metadata["overlay_id"],
"secret_ref": metadata["secret_ref"],
"expires_at": metadata["expires_at"],
"binding": expected,
}
token_text = secrets.token_urlsafe(48)
token_bytes = token_text.encode("ascii")
digest = _token_hash(token_bytes)
self._assert_hash_unique(digest)
overlay_id = "overlay-" + hashlib.sha256(
(
f"{self.config.binding.task_id}\0"
f"{self.config.binding.base_oid}\0"
f"{self.config.binding.target_branch}\0"
f"{digest}"
).encode("utf-8")
).hexdigest()[:32]
issued_at = utc_now()
expires_at = issued_at + timedelta(seconds=ttl_seconds)
overlay_dir = self.config.state_root / overlay_id
secret_dir = self.config.secret_root / overlay_id
_prepare_directory(overlay_dir)
_prepare_directory(secret_dir)
secret_ref = secret_dir / "bearer.token"
_atomic_write(secret_ref, token_bytes + b"\n", 0o600)
metadata: dict[str, Any] = {
"schema": OVERLAY_SCHEMA,
"overlay_id": overlay_id,
"status": "ACTIVE",
**self.config.binding.as_dict(),
"owner_identity": self.config.owner_identity,
"role": "owner",
"scope": OVERLAY_SCOPE,
"wildcard_scope": False,
"one_time_use": True,
"token_sha256": digest,
"secret_ref": str(secret_ref),
"authorization_assertion_ref": authorization_assertion_ref,
"operator_identity": operator_identity,
"issued_at": iso_utc(issued_at),
"expires_at": iso_utc(expires_at),
"attempt_count": 0,
}
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
metadata_path = overlay_dir / "metadata.json"
_atomic_write_json(metadata_path, metadata)
_atomic_write_json(
overlay_dir / "audit-issued.json",
{
"event": "TASK_IDENTITY_OVERLAY_ISSUED",
"overlay_id": overlay_id,
"task_id": self.config.binding.task_id,
"binding_digest_sha256": sha256_json(
self.config.binding.as_dict()
),
"authorization_assertion_ref": authorization_assertion_ref,
"operator_identity": operator_identity,
"issued_at": metadata["issued_at"],
"expires_at": metadata["expires_at"],
"token_sha256": digest,
},
)
return {
"status": "OVERLAY_ACTIVE",
"idempotent": False,
"overlay_id": overlay_id,
"secret_ref": str(secret_ref),
"expires_at": metadata["expires_at"],
"binding": self.config.binding.as_dict(),
}
def authenticate_and_lock(
self, token: str, request: Mapping[str, Any]
) -> OverlaySession:
if not token or len(token) > 4096:
raise IdentityOverlayError("OVERLAY_AUTHENTICATION_REQUIRED", 401)
digest = _token_hash(token.encode("utf-8"))
with self._lock():
matches: list[tuple[Path, dict[str, Any]]] = []
for path in self._metadata_paths():
metadata = _read_json(path)
_verify_metadata(metadata)
if metadata.get("token_sha256") == digest:
matches.append((path, metadata))
if not matches:
raise IdentityOverlayError("OVERLAY_CREDENTIAL_INVALID", 401)
if len(matches) != 1:
raise IdentityOverlayError("OVERLAY_SECRET_IDENTITY_COLLISION", 409)
path, metadata = matches[0]
if metadata.get("status") in TERMINAL_STATES:
raise IdentityOverlayError("OVERLAY_CREDENTIAL_INACTIVE", 401)
if _parse_utc(str(metadata["expires_at"])) <= utc_now():
metadata["status"] = "EXPIRED"
metadata["expired_at"] = iso_utc(utc_now())
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
_atomic_write_json(path, metadata)
_safe_unlink(Path(str(metadata["secret_ref"])))
raise IdentityOverlayError("OVERLAY_CREDENTIAL_EXPIRED", 401)
expected = self.config.binding.as_dict()
for key, value in expected.items():
if metadata.get(key) != value:
raise IdentityOverlayError(
"OVERLAY_METADATA_BINDING_MISMATCH", 409
)
if metadata.get("owner_identity") != self.config.owner_identity:
raise IdentityOverlayError("OVERLAY_OWNER_IDENTITY_MISMATCH", 409)
if metadata.get("scope") != OVERLAY_SCOPE:
raise IdentityOverlayError("OVERLAY_SCOPE_MISMATCH", 409)
if metadata.get("wildcard_scope") is not False:
raise IdentityOverlayError("OVERLAY_WILDCARD_SCOPE_REJECTED", 409)
# Validate detailed request policy only after the bearer identity is
# proven, so unauthenticated callers cannot probe task bindings.
validated = _validate_request_binding(
request, self.config.binding, self.config.max_ttl_seconds
)
request_digest = sha256_json(validated)
previous = str(metadata.get("request_digest_sha256") or "")
if metadata.get("status") == "LOCKED" and previous != request_digest:
raise IdentityOverlayError("OVERLAY_REPLAY_REQUEST_MISMATCH", 409)
metadata["status"] = "LOCKED"
metadata["request_digest_sha256"] = request_digest
metadata["locked_at"] = metadata.get("locked_at") or iso_utc(utc_now())
metadata["attempt_count"] = int(metadata.get("attempt_count") or 0) + 1
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
_atomic_write_json(path, metadata)
return OverlaySession(
overlay_id=str(metadata["overlay_id"]),
metadata_path=path,
secret_ref=Path(str(metadata["secret_ref"])),
owner_identity=str(metadata["owner_identity"]),
request_digest_sha256=request_digest,
binding=self.config.binding,
)
def complete(
self, session: OverlaySession, result: Mapping[str, Any]
) -> dict[str, Any]:
with self._lock():
metadata = _read_json(session.metadata_path)
_verify_metadata(metadata)
if metadata.get("status") == "CONSUMED":
return metadata
if metadata.get("status") != "LOCKED":
raise IdentityOverlayError("OVERLAY_NOT_LOCKED", 409)
if (
metadata.get("request_digest_sha256")
!= session.request_digest_sha256
):
raise IdentityOverlayError("OVERLAY_REQUEST_DIGEST_MISMATCH", 409)
metadata["status"] = "CONSUMED"
metadata["consumed_at"] = iso_utc(utc_now())
metadata["result_digest_sha256"] = sha256_json(dict(result))
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
_atomic_write_json(session.metadata_path, metadata)
_safe_unlink(session.secret_ref)
_atomic_write_json(
session.metadata_path.parent / "audit-consumed.json",
{
"event": "TASK_IDENTITY_OVERLAY_CONSUMED",
"overlay_id": session.overlay_id,
"task_id": session.binding.task_id,
"request_digest_sha256": session.request_digest_sha256,
"result_digest_sha256": metadata["result_digest_sha256"],
"consumed_at": metadata["consumed_at"],
"secret_removed": True,
},
)
return metadata
def fail(
self,
session: OverlaySession,
*,
error_code: str,
error_status: int,
) -> dict[str, Any]:
with self._lock():
metadata = _read_json(session.metadata_path)
_verify_metadata(metadata)
if metadata.get("status") in TERMINAL_STATES:
return metadata
metadata["status"] = "FAILED"
metadata["failed_at"] = iso_utc(utc_now())
metadata["error_code"] = error_code
metadata["error_status"] = int(error_status)
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
_atomic_write_json(session.metadata_path, metadata)
_safe_unlink(session.secret_ref)
_atomic_write_json(
session.metadata_path.parent / "audit-failed.json",
{
"event": "TASK_IDENTITY_OVERLAY_FAILED",
"overlay_id": session.overlay_id,
"task_id": session.binding.task_id,
"request_digest_sha256": session.request_digest_sha256,
"error_code": error_code,
"error_status": int(error_status),
"failed_at": metadata["failed_at"],
"secret_removed": True,
},
)
return metadata
def revoke(self, overlay_id: str, *, reason: str) -> dict[str, Any]:
if self.config.require_root_issue and os.geteuid() != 0:
raise IdentityOverlayError("OVERLAY_ROOT_ISSUER_REQUIRED", 403)
if not OVERLAY_ID_RE.fullmatch(overlay_id):
raise IdentityOverlayError("OVERLAY_ID_INVALID")
if not AUTH_ASSERTION_RE.fullmatch(reason):
raise IdentityOverlayError("OVERLAY_REVOCATION_REASON_INVALID")
with self._lock():
path = self.config.state_root / overlay_id / "metadata.json"
metadata = _read_json(path)
_verify_metadata(metadata)
if metadata.get("status") not in TERMINAL_STATES:
metadata["status"] = "REVOKED"
metadata["revoked_at"] = iso_utc(utc_now())
metadata["revocation_reason"] = reason
metadata["metadata_digest_sha256"] = _metadata_digest(metadata)
_atomic_write_json(path, metadata)
_safe_unlink(Path(str(metadata["secret_ref"])))
return metadata
def status(self, overlay_id: str) -> dict[str, Any]:
if not OVERLAY_ID_RE.fullmatch(overlay_id):
raise IdentityOverlayError("OVERLAY_ID_INVALID")
metadata = _read_json(
self.config.state_root / overlay_id / "metadata.json"
)
_verify_metadata(metadata)
return {
key: value
for key, value in metadata.items()
if key not in {"token_sha256"}
}
class TaskIdentityOverlayService:
"""Authenticate one overlay credential and invoke the existing broker."""
def __init__(
self,
config: IdentityOverlayConfig,
provisioner: TaskWriterProvisioner | None = None,
) -> None:
self.config = config
self.store = TaskIdentityOverlayStore(config)
broker_state = config.broker_state_root or (
config.state_root / "owner-authorization-events"
)
self.broker = OwnerAuthorizationBroker(
OwnerAuthorizationBrokerConfig(
repository_identity=config.binding.repository,
state_root=broker_state,
allowed_owner_identities=frozenset({config.owner_identity}),
allowed_intake_scopes=frozenset({OVERLAY_SCOPE}),
allowed_credential_ids=frozenset(),
max_expiry_seconds=config.max_ttl_seconds,
),
provisioner=provisioner,
)
def provision(
self, request: Mapping[str, Any], bearer_token: str
) -> dict[str, Any]:
session = self.store.authenticate_and_lock(bearer_token, request)
principal = OwnerPrincipal(
identity=session.owner_identity,
role="owner",
authenticated_scope=OVERLAY_SCOPE,
token_scopes=frozenset({OVERLAY_SCOPE}),
credential_id=session.overlay_id,
)
try:
result = self.broker.authorize_and_provision(request, principal)
if result.get("status") != "READY_FOR_CODE_AUTHORING":
raise IdentityOverlayError("OVERLAY_PROVISIONER_NOT_READY", 409)
self.store.complete(session, result)
response = dict(result)
response.update(
{
"identity_overlay_id": session.overlay_id,
"identity_overlay_consumed": True,
"identity_overlay_secret_removed": True,
"identity_overlay_scope": OVERLAY_SCOPE,
}
)
return response
except (OwnerAuthorizationError, ProvisioningError) as exc:
self.store.fail(
session, error_code=exc.code, error_status=exc.status
)
raise
except IdentityOverlayError as exc:
self.store.fail(
session, error_code=exc.code, error_status=exc.status
)
raise
except Exception as exc:
self.store.fail(
session,
error_code="IDENTITY_OVERLAY_INTERNAL_ERROR",
error_status=500,
)
raise IdentityOverlayError(
"IDENTITY_OVERLAY_INTERNAL_ERROR", 500
) from exc
+21
View File
@@ -0,0 +1,21 @@
"""Task Plane runtime package."""
from .models import (
EvidenceEnvelope,
ProbeEvidence,
ProbeOutcome,
TaskPlan,
TaskRecord,
TaskStatus,
TaskStateTransitionError,
)
__all__ = [
"EvidenceEnvelope",
"ProbeEvidence",
"ProbeOutcome",
"TaskPlan",
"TaskRecord",
"TaskStatus",
"TaskStateTransitionError",
]
@@ -0,0 +1,17 @@
"""Adapters for executing tasks via worker backends."""
from .base import WorkerAdapter
from .openclaw import OpenClawAdapter
from .codex import CodexAdapter
from .hermes import HermesAdapter
from .a2a import A2AAdapter
from .systemd_worker import SystemdWorkerAdapter
__all__ = [
"WorkerAdapter",
"OpenClawAdapter",
"CodexAdapter",
"HermesAdapter",
"A2AAdapter",
"SystemdWorkerAdapter",
]
@@ -0,0 +1,31 @@
from __future__ import annotations
from .base import WorkerAdapter
class A2AAdapter(WorkerAdapter):
worker_name = "a2a"
def capabilities(self) -> list[str]:
return ["task.exchange", "agent.handoff"]
def claim(self, task):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "claimed"}
def execute(self, task, claim):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "SHADOW_VERIFIED", "claim_id": claim.claim_id, "execution_mode": "shadow", "production_side_effects": False}
def heartbeat(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "alive"}
def report_progress(self, claim, progress):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "progress": progress}
def probe(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": claim.status}
def cancel(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "cancelled"}
def collect_result(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "result": "not_collected"}
@@ -0,0 +1,41 @@
"""Base worker adapter contract."""
from __future__ import annotations
import abc
class WorkerAdapter(abc.ABC):
worker_name = "base"
@abc.abstractmethod
def capabilities(self) -> list[str]:
raise NotImplementedError
@abc.abstractmethod
def claim(self, task):
raise NotImplementedError
@abc.abstractmethod
def execute(self, task, claim):
raise NotImplementedError
@abc.abstractmethod
def heartbeat(self, claim):
raise NotImplementedError
@abc.abstractmethod
def report_progress(self, claim, progress):
raise NotImplementedError
@abc.abstractmethod
def probe(self, claim):
raise NotImplementedError
@abc.abstractmethod
def cancel(self, claim):
raise NotImplementedError
@abc.abstractmethod
def collect_result(self, claim):
raise NotImplementedError
@@ -0,0 +1,31 @@
from __future__ import annotations
from .base import WorkerAdapter
class CodexAdapter(WorkerAdapter):
worker_name = "codex"
def capabilities(self) -> list[str]:
return ["code.read", "code.write.patch", "test.python", "review.security"]
def claim(self, task):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "claimed"}
def execute(self, task, claim):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "SHADOW_VERIFIED", "claim_id": claim.claim_id, "execution_mode": "shadow", "production_side_effects": False}
def heartbeat(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "alive"}
def report_progress(self, claim, progress):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "progress": progress}
def probe(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": claim.status}
def cancel(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "cancelled"}
def collect_result(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "result": "not_collected"}
@@ -0,0 +1,31 @@
from __future__ import annotations
from .base import WorkerAdapter
class HermesAdapter(WorkerAdapter):
worker_name = "hermes"
def capabilities(self) -> list[str]:
return ["architecture.review", "policy.review", "threat.model"]
def claim(self, task):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "claimed"}
def execute(self, task, claim):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "SHADOW_VERIFIED", "claim_id": claim.claim_id, "execution_mode": "shadow", "production_side_effects": False}
def heartbeat(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "alive"}
def report_progress(self, claim, progress):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "progress": progress}
def probe(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": claim.status}
def cancel(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "cancelled"}
def collect_result(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "result": "not_collected"}
@@ -0,0 +1,31 @@
from __future__ import annotations
from .base import WorkerAdapter
class OpenClawAdapter(WorkerAdapter):
worker_name = "openclaw"
def capabilities(self) -> list[str]:
return ["code.read", "code.write.patch", "test.execute", "runtime.shadow"]
def claim(self, task):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "claimed"}
def execute(self, task, claim):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "SHADOW_VERIFIED", "claim_id": claim.claim_id, "execution_mode": "shadow", "production_side_effects": False}
def heartbeat(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "alive"}
def report_progress(self, claim, progress):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "progress": progress}
def probe(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": claim.status}
def cancel(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "cancelled"}
def collect_result(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "result": "not_collected"}
@@ -0,0 +1,31 @@
from __future__ import annotations
from .base import WorkerAdapter
class SystemdWorkerAdapter(WorkerAdapter):
worker_name = "systemd_worker"
def capabilities(self) -> list[str]:
return ["systemd.exec", "runtime.shell"]
def claim(self, task):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "claimed"}
def execute(self, task, claim):
return {"worker": self.worker_name, "task_id": task.task_id, "status": "SHADOW_VERIFIED", "claim_id": claim.claim_id, "execution_mode": "shadow", "production_side_effects": False}
def heartbeat(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "alive"}
def report_progress(self, claim, progress):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "progress": progress}
def probe(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": claim.status}
def cancel(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "status": "cancelled"}
def collect_result(self, claim):
return {"worker": self.worker_name, "claim_id": claim.claim_id, "result": "not_collected"}
@@ -0,0 +1,27 @@
"""Agent registry abstraction."""
from __future__ import annotations
from .errors import NotFoundError
class AgentRegistry:
def __init__(self):
self._agents: dict[str, dict] = {}
def register(self, agent_id: str, metadata: dict) -> None:
self._agents[agent_id] = {"agent_id": agent_id, **metadata}
def unregister(self, agent_id: str) -> None:
if agent_id not in self._agents:
raise NotFoundError("agent")
del self._agents[agent_id]
def list(self) -> list[dict]:
return list(self._agents.values())
def get(self, agent_id: str) -> dict:
try:
return self._agents[agent_id]
except KeyError:
raise NotFoundError("agent")
+557
View File
@@ -0,0 +1,557 @@
"""Task Plane HTTP API implementation (stdlib-only)."""
from __future__ import annotations
import json
import os
import time
import uuid
import urllib.parse
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from typing import Any, Callable
from .delegation import DelegationEngine
from .errors import (
ConflictError,
ForbiddenError,
NotFoundError,
PayloadTooLargeError,
TaskPlaneHTTPError,
UnauthorizedError,
ValidationError,
)
from .evidence import EvidenceService
from .journal import TaskJournal
from .lease import LeaseManager
from .models import (
EvidenceEnvelope,
ResultEnvelope,
TaskPlan,
TaskRecord,
iso8601_now,
contains_secret_text,
validate_identifier,
)
from .policy import PolicyEngine
from .store import TaskPlaneStore
from .worker_registry import WorkerRegistry
def _json_payload(data: dict[str, Any], status: int = 200) -> tuple[int, bytes]:
body = json.dumps(data, ensure_ascii=False).encode("utf-8")
return status, body
def _bounded_int(value: Any, name: str, minimum: int, maximum: int) -> int:
try:
number = int(value)
except (TypeError, ValueError) as exc:
raise ValidationError(f"invalid {name}") from exc
if not minimum <= number <= maximum:
raise ValidationError(f"invalid {name}")
return number
class TaskPlaneService:
def __init__(self, store: TaskPlaneStore, policy: PolicyEngine | None = None):
self.store = store
self.policy = policy or PolicyEngine(
execution_enabled=False,
identity_capabilities={
"anon": {"task.creator", "task.read", "task.audit"}
},
)
self.journal = TaskJournal(store)
self.lease = LeaseManager(store)
self.workers = WorkerRegistry(store)
self.evidence = EvidenceService(store)
self.delegation = DelegationEngine(store)
self.store.probe_attestation_verifier = self.policy.probe_attestation_verifier
def create_task(self, body: dict[str, Any], idempotency_key: str | None = None) -> dict[str, Any]:
if body.get("parent_task_id"):
raise ForbiddenError("child tasks must use the delegate endpoint")
issuer = body.get("issuer", {})
actor = issuer.get("id", "anon") if isinstance(issuer, dict) else "anon"
self.policy.assert_allowed("create", actor)
try:
plan = TaskPlan.from_dict(body)
except (TypeError, ValueError) as exc:
raise ValidationError("invalid task plan") from exc
task = self.store.create_task_with_event(plan, idempotency_key=idempotency_key)
return task.to_dict()
def list_tasks(self, actor: str | None = None) -> list[dict[str, Any]]:
if actor is not None:
self.policy.assert_allowed("read", actor)
return [task.to_dict(include_task=False) for task in self.store.list_tasks()]
def get_task(self, task_id: str, actor: str | None = None) -> dict[str, Any]:
if actor is not None:
self.policy.assert_allowed("read", actor, task=self.store.get_task(task_id))
return self.store.get_task(task_id).to_dict()
def get_worker(self, worker_id: str, actor: str | None = None) -> dict[str, Any]:
if actor is not None:
self.policy.assert_allowed("read", actor)
return self.workers.get(worker_id).to_dict()
def list_workers(self, actor: str | None = None) -> list[dict[str, Any]]:
if actor is not None:
self.policy.assert_allowed("read", actor)
return [w.to_dict() for w in self.workers.list()]
def register_worker(self, body: dict[str, Any]) -> dict[str, Any]:
actor = str(body.get("actor_identity") or body.get("identity") or "")
is_admin = actor in self.policy.admin_identities
try:
existing = self.store.get_worker(str(body.get("worker_id") or ""))
except NotFoundError:
existing = None
if existing is not None and str(body.get("identity") or "") != existing.identity:
raise ConflictError("worker identity takeover rejected")
self.policy.assert_readonly_worker_registration(actor, body)
self.policy.assert_allowed(
"worker.admin" if is_admin else "worker.register", actor
)
try:
worker = self.workers.register(
body,
actor_identity=actor,
is_admin=is_admin,
expected_revision=body.get("expected_revision"),
)
except (TypeError, ValueError, KeyError) as exc:
raise ValidationError("invalid worker registration") from exc
return worker.to_dict()
def worker_heartbeat(self, worker_id: str, payload: dict[str, Any]) -> dict[str, Any]:
interval = payload.get("heartbeat_interval_seconds")
actor = str(payload.get("actor_identity") or "")
self.policy.assert_readonly_worker_binding(actor, worker_id)
self.policy.assert_allowed("worker.register", actor)
return self.workers.heartbeat(
worker_id,
heartbeat_interval_seconds=interval,
actor_identity=actor,
expected_revision=payload.get("expected_revision"),
).to_dict()
def disable_worker(self, worker_id: str, payload: dict[str, Any]) -> dict[str, Any]:
actor = str(payload.get("actor_identity") or "")
self.policy.assert_allowed("worker.admin", actor)
return self.store.disable_worker(
worker_id,
actor_identity=actor,
expected_revision=int(payload.get("expected_revision")),
reason=str(payload.get("reason") or "disabled by administrator"),
).to_dict()
def restore_worker(self, worker_id: str, payload: dict[str, Any]) -> dict[str, Any]:
actor = str(payload.get("actor_identity") or "")
self.policy.assert_allowed("worker.admin", actor)
return self.store.restore_worker(
worker_id,
actor_identity=actor,
expected_revision=int(payload.get("expected_revision")),
).to_dict()
def claim(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
actor = body.get("actor_identity") or body.get("worker_identity") or "worker"
self.policy.assert_allowed("claim", actor)
worker_id = body.get("worker_id")
if not worker_id:
raise ValidationError("worker_id required")
task = self.store.get_task(task_id)
self.policy.assert_readonly_worker_binding(actor, str(worker_id))
self.policy.assert_allowed("claim", actor, task=task)
lease_seconds = _bounded_int(
body.get("lease_seconds", 30), "lease_seconds", 1, 3600
)
claim = self.lease.claim(
task_id=task_id,
worker_id=worker_id,
worker_identity=actor,
lease_seconds=lease_seconds,
expected_revision=task.revision,
operation_digest=task.task_plan.operation_digest(),
)
return {"claim_id": claim.claim_id, **claim.to_dict()}
def heartbeat(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
claim_id = body.get("claim_id")
if not claim_id:
raise ValidationError("claim_id required")
actor = body.get("actor_identity") or body.get("worker_identity") or "worker"
task = self.store.get_task(task_id)
claim = self.store.get_claim(claim_id)
self.policy.assert_readonly_worker_binding(actor, claim.worker_id)
self.policy.assert_allowed("heartbeat", actor, task=task)
if claim.worker_identity != actor:
raise ForbiddenError("claim identity mismatch")
lease_seconds = _bounded_int(
body.get("lease_seconds", 30), "lease_seconds", 1, 3600
)
claim = self.lease.heartbeat(task_id, claim_id, lease_seconds)
return {"claim_id": claim_id, "lease_until": claim.lease_until}
def progress(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
claim_id = body.get("claim_id")
if not claim_id:
raise ValidationError("claim_id required")
actor = body.get("actor_identity") or body.get("worker_identity") or "worker"
progress = body.get("progress")
stage = body.get("stage", "RUNNING")
message = body.get("message")
if progress is None:
raise ValidationError("progress required")
task = self.store.get_task(task_id)
claim = self.store.get_claim(claim_id)
self.policy.assert_readonly_worker_binding(actor, claim.worker_id)
self.policy.assert_allowed("progress", actor, task=task)
if task.task_plan.data_classification == "credential-secret":
if message not in {None, ""} or contains_secret_text(stage):
raise ValidationError(
"credential-secret progress must contain metadata only"
)
lease_seconds = _bounded_int(
body.get("lease_seconds", 30), "lease_seconds", 1, 3600
)
claim = self.store.record_progress(
task_id,
claim_id,
actor,
progress=progress,
stage=str(stage),
message=None if message is None else str(message),
lease_seconds=lease_seconds,
)
return {"claim_id": claim.claim_id, "status": claim.status, "progress": progress}
def complete(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
claim_id = body.get("claim_id")
if not claim_id:
raise ValidationError("claim_id required")
actor = body.get("actor_identity") or body.get("worker_identity") or "worker"
task = self.store.get_task(task_id)
status = body.get("status", "COMPLETED")
outputs = body.get("outputs", [])
evidences = body.get("evidence", [])
if not isinstance(outputs, list):
raise ValidationError("outputs must be list")
if not isinstance(evidences, list):
raise ValidationError("evidence must be list")
if any(not isinstance(ev, dict) for ev in evidences):
raise ValidationError("evidence items must be objects")
claim = self.store.get_claim(claim_id)
self.policy.assert_readonly_worker_binding(actor, claim.worker_id)
self.policy.assert_allowed("result", actor, task=task)
if claim.worker_identity != actor:
raise ForbiddenError("claim identity mismatch")
if task.task_plan.data_classification == "credential-secret":
allowed_result_keys = {
"kind",
"hash",
"sha256",
"count",
"status",
"error_code",
"signature",
"attestation_ref",
}
for entry in [*outputs, *evidences]:
if (
not isinstance(entry, dict)
or set(entry) - allowed_result_keys
or contains_secret_text(entry)
):
raise ValidationError(
"credential-secret result must contain hash/count/status only"
)
result = ResultEnvelope(
task_id=task_id,
claim_id=claim_id,
worker_identity=actor,
status=status,
outputs=outputs,
evidence=evidences,
started_at=str(body.get("started_at") or iso8601_now()),
completed_at=str(body.get("completed_at") or iso8601_now()),
result_digest=body.get("result_digest"),
signature=body.get("signature"),
attestation_ref=body.get("attestation_ref"),
operation_digest=claim.operation_digest,
)
evidence_decision = self.policy.evidence_verifier.verify(task, evidences)
if not evidence_decision.allowed:
raise ForbiddenError(evidence_decision.reason)
result_decision = self.policy.verify_result(result, task)
if not result_decision.allowed:
raise ForbiddenError(result_decision.reason)
result.result_digest = result.computed_digest()
previous = self.store._last_evidence_hash(task_id)
evidence_entries = []
for evidence in evidences:
envelope = EvidenceEnvelope.create(
task_id=task_id,
claim_id=claim_id,
kind=str(evidence.get("kind") or evidence.get("type") or "generic"),
worker_identity=actor,
payload=evidence,
previous_evidence_hash=previous,
signature=evidence.get("signature"),
attestation_ref=evidence.get("attestation_ref"),
)
evidence_entries.append(envelope)
previous = envelope.evidence_digest
self.store.complete_claim_with_evidence_and_event(
task_id,
claim_id,
status
if status
in {"COMPLETED", "FAILED_FINAL", "FAILED_RECOVERABLE", "CANCELLED"}
else "COMPLETED",
result.to_dict(),
evidence_entries,
)
return result.to_dict()
def fail(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
claim_id = body.get("claim_id")
if not claim_id:
raise ValidationError("claim_id required")
reason = body.get("reason", "failed")
actor = body.get("actor_identity") or body.get("worker_identity") or "worker"
self.policy.assert_allowed("fail", actor)
task = self.store.get_task(task_id)
self.policy.assert_allowed("fail", actor, task=task)
claim = self.store.get_claim(claim_id)
if claim.worker_identity != actor:
raise ForbiddenError("claim identity mismatch")
status = "FAILED_RECOVERABLE" if body.get("recoverable", False) else "FAILED_FINAL"
self.store.fail_claim_with_event(task_id, claim_id, status, str(reason))
return {"task_id": task_id, "status": status, "reason": reason}
def cancel(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
actor = body.get("actor_identity") or body.get("worker_identity") or "owner"
self.policy.assert_allowed("cancel", actor)
reason = body.get("reason", "cancelled")
task = self.store.get_task(task_id)
self.policy.assert_allowed("cancel", actor, task=task)
updated = self.store.cancel_task_with_event(
task_id, task.revision, reason=reason
)
return updated.to_dict()
def delegate(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
actor = body.get("actor_identity") or "owner"
self.policy.assert_allowed("delegate", actor)
parent = self.store.get_task(task_id)
self.policy.assert_allowed("delegate", actor, task=parent)
if "task" in body:
try:
child_plan = TaskPlan.from_dict(body["task"])
except (TypeError, ValueError) as exc:
raise ValidationError("invalid child task plan") from exc
else:
plan_body = dict(body)
plan_body.pop("actor_identity", None)
try:
child_plan = TaskPlan.from_dict(plan_body)
except (TypeError, ValueError) as exc:
raise ValidationError("invalid child task plan") from exc
child_plan.parent_task_id = task_id
child = self.delegation.delegate(task_id, child_plan, caller=actor)
return child.to_dict()
def get_events(self, task_id: str, actor: str | None = None) -> list[dict[str, Any]]:
if actor is not None:
self.policy.assert_allowed("audit", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
return self.store.list_events(task_id)
def get_evidence(self, task_id: str, actor: str | None = None) -> list[dict[str, Any]]:
if actor is not None:
self.policy.assert_allowed("audit", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
return self.evidence.list(task_id)
def get_children(self, task_id: str, actor: str | None = None) -> list[str]:
if actor is not None:
self.policy.assert_allowed("read", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
return self.store.list_children(task_id)
class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
service: TaskPlaneService | None = None
authenticator: Callable[[str], str | None] | None = None
max_body_bytes = 1_000_000
def log_message(self, format: str, *args: Any) -> None:
# BaseHTTPRequestHandler logs the raw request target; suppress it so
# attacker-controlled paths or query strings cannot leak credentials.
return
def _json(self) -> tuple[str | None, dict[str, Any]]:
try:
length = int(self.headers.get("Content-Length", "0"))
except ValueError as exc:
raise ValidationError("invalid content length") from exc
if length < 0:
raise ValidationError("invalid content length")
if length > self.max_body_bytes:
raise PayloadTooLargeError("payload exceeds limit")
raw = self.rfile.read(length) if length > 0 else b""
if not raw:
return None, {}
try:
decoded = raw.decode("utf-8")
payload = json.loads(decoded)
if not isinstance(payload, dict):
raise ValidationError("JSON payload must be an object")
return decoded, payload
except (UnicodeDecodeError, json.JSONDecodeError) as exc:
raise ValidationError("invalid JSON payload") from exc
def _response(self, status: int, payload: Any) -> None:
encoded = json.dumps(payload, ensure_ascii=False).encode("utf-8")
self.send_response(status)
self.send_header("Content-Type", "application/json; charset=utf-8")
self.send_header("Content-Length", str(len(encoded)))
self.end_headers()
self.wfile.write(encoded)
def _authenticated_identity(self) -> str:
authenticator = type(self).authenticator
if authenticator is None:
raise UnauthorizedError("authentication unavailable")
header = self.headers.get("Authorization", "")
if not header.startswith("Bearer "):
raise UnauthorizedError("bearer authentication required")
token = header[7:]
if not token or any(ord(ch) < 33 for ch in token):
raise UnauthorizedError("invalid bearer token")
identity = authenticator(token)
if not identity:
raise UnauthorizedError("invalid credentials")
return str(identity)
def _handle(self, handler):
try:
if self.service is None:
raise TaskPlaneHTTPError("service not configured")
result = handler()
if isinstance(result, tuple):
status, payload = result
else:
status, payload = 200, result
self._response(status, payload)
except TaskPlaneHTTPError as exc:
self._response(exc.status_code, exc.to_payload())
except Exception as exc: # defensive
if isinstance(exc, TaskPlaneHTTPError):
raise
self._response(
500,
{"error": "internal_error", "message": "internal server error"},
)
def do_GET(self): # noqa: N802
def route():
path = urllib.parse.urlparse(self.path).path
parts = [p for p in path.split("/") if p]
for part in parts[1:]:
if part not in {"events", "evidence", "children", "heartbeat"}:
try:
validate_identifier(part, "resource identifier")
except ValueError as exc:
raise ValidationError("invalid resource identifier") from exc
identity = self._authenticated_identity()
if len(parts) == 1 and parts[0] == "tasks":
return 200, self.service.list_tasks(identity)
if len(parts) == 2 and parts[0] == "tasks":
task_id = parts[1]
return 200, self.service.get_task(task_id, identity)
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "events":
return 200, self.service.get_events(parts[1], identity)
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "evidence":
return 200, self.service.get_evidence(parts[1], identity)
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "children":
return 200, self.service.get_children(parts[1], identity)
if len(parts) == 1 and parts[0] == "workers":
return 200, self.service.list_workers(identity)
if len(parts) == 2 and parts[0] == "workers":
return 200, self.service.get_worker(parts[1], identity)
raise NotFoundError("unknown endpoint")
self._handle(route)
def do_POST(self): # noqa: N802
def route():
body_text, body = self._json()
identity = self._authenticated_identity()
path = urllib.parse.urlparse(self.path).path
parts = [p for p in path.split("/") if p]
headers = {k.lower(): v for k, v in self.headers.items()}
if len(parts) == 1 and parts[0] == "tasks":
issuer = body.get("issuer")
if issuer and (
not isinstance(issuer, dict) or issuer.get("id") != identity
):
raise ForbiddenError("issuer identity mismatch")
body["issuer"] = {"type": "authenticated", "id": identity}
return 201, self.service.create_task(body or {}, headers.get("idempotency-key"))
if body.get("actor_identity") not in {None, identity}:
raise ForbiddenError("actor identity mismatch")
body["actor_identity"] = identity
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "claim":
return 200, self.service.claim(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "heartbeat":
return 200, self.service.heartbeat(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "progress":
return 200, self.service.progress(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "result":
return 200, self.service.complete(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "fail":
return 200, self.service.fail(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "cancel":
return 200, self.service.cancel(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "delegate":
return 200, self.service.delegate(parts[1], body or {})
if len(parts) == 2 and parts[0] == "workers" and parts[1] == "register":
if (
body.get("identity") != identity
and identity not in self.service.policy.admin_identities
):
raise ForbiddenError("worker identity mismatch")
return 201, self.service.register_worker(body or {})
if len(parts) == 3 and parts[0] == "workers" and parts[2] == "heartbeat":
return 200, self.service.worker_heartbeat(parts[1], body or {})
if len(parts) == 3 and parts[0] == "workers" and parts[2] == "disable":
return 200, self.service.disable_worker(parts[1], body or {})
if len(parts) == 3 and parts[0] == "workers" and parts[2] == "restore":
return 200, self.service.restore_worker(parts[1], body or {})
raise NotFoundError("unknown endpoint")
self._handle(route)
def run_server(
host: str,
port: int,
service: TaskPlaneService,
*,
authenticator: Callable[[str], str | None] | None = None,
):
handler = TaskPlaneRequestHandler
handler.service = service
handler.authenticator = authenticator
server = ThreadingHTTPServer((host, port), handler)
server.serve_forever()
@@ -0,0 +1,36 @@
"""Capability matcher helpers."""
from __future__ import annotations
from fnmatch import fnmatch
def _normalize_capability(capability: str) -> str:
return capability.strip().lower()
def capability_matches(granted: str, required: str) -> bool:
granted = _normalize_capability(granted)
required = _normalize_capability(required)
if granted == required:
return True
# A granted wildcard such as "test.*" should match required "test.unit"
if "*" in granted:
return fnmatch(required, granted)
# A required wildcard is treated as a suffix/prefix requirement for callers.
if "*" in required:
return fnmatch(granted, required)
return False
def has_capabilities(granted_caps: list[str], required_caps: list[str]) -> bool:
granted = [_normalize_capability(c) for c in granted_caps]
for required in required_caps:
required = _normalize_capability(required)
if not any(capability_matches(g, required) for g in granted):
return False
return True
def normalize_capabilities(caps: list[str]) -> list[str]:
return sorted({_normalize_capability(cap) for cap in caps})
@@ -0,0 +1,87 @@
"""Delegation guardrails."""
from __future__ import annotations
from .errors import ConflictError, ForbiddenError, NotFoundError
from .models import Budget, TaskPlan, TaskRecord
from .store import TaskPlaneStore
class DelegationEngine:
def __init__(self, store: TaskPlaneStore):
self.store = store
def can_delegate(self, parent_id: str, child_plan: TaskPlan) -> bool:
parent = self.store.get_task(parent_id)
if parent.status == "COMPLETED":
raise ConflictError("cannot delegate completed task")
if parent.task_plan.max_delegation_depth <= 0:
raise ConflictError("max delegation depth not configured")
depth = self._depth(parent_id)
if depth >= parent.task_plan.max_delegation_depth:
raise ConflictError("max delegation depth reached")
if child_plan.parent_task_id not in {None, parent_id}:
raise ConflictError("delegation parent mismatch")
current = parent_id
while current:
if child_plan.task_id and current == child_plan.task_id:
raise ConflictError("delegation cycle")
current = self.store.parent_of(current)
return True
def delegate(self, parent_id: str, child_plan: TaskPlan, *, caller: str) -> TaskRecord:
self.can_delegate(parent_id, child_plan)
parent = self.store.get_task(parent_id)
if child_plan.required_capabilities and not set(child_plan.required_capabilities).issubset(
set(parent.task_plan.required_capabilities or [])
):
raise ForbiddenError("child capabilities exceed parent constraints")
risk_rank = {"low": 1, "medium": 2, "high": 3}
if risk_rank.get(child_plan.risk, 99) > risk_rank.get(parent.task_plan.risk, -1):
raise ForbiddenError("child risk exceeds parent constraints")
child_plan.parent_task_id = parent_id
child_plan.correlation_id = parent.task_plan.correlation_id or parent.task_id
child_plan.preferred_workers = parent.task_plan.preferred_workers
child_plan.max_delegation_depth = min(
child_plan.max_delegation_depth,
parent.task_plan.max_delegation_depth,
)
child_plan.approval_policy = parent.task_plan.approval_policy
parent_budget = self.store.get_task_budget(parent_id)
required = Budget()
for name in ("time_seconds", "attempts", "tokens", "cost"):
available = getattr(parent_budget, name)
requested = getattr(child_plan.budgets, name)
if name == "attempts" and requested is None:
requested = 1
if available is not None:
requested = 0 if requested is None else requested
if requested > available:
raise ConflictError(f"budget exceeded for {name}")
setattr(child_plan.budgets, name, requested)
setattr(required, name, requested)
# A retried identical delegation is a read, not a second budget charge.
if child_plan.task_id:
try:
existing = self.store.get_task(child_plan.task_id)
except NotFoundError:
existing = None
if existing is not None:
if (
existing.task_plan.parent_task_id == parent_id
and existing.task_plan.operation_digest()
== child_plan.operation_digest()
):
return existing
raise ConflictError("delegation task_id conflict")
return self.store.delegate_task_with_budget_and_events(
parent_id, child_plan, caller=caller
)
def _depth(self, task_id: str) -> int:
return self.store._compute_depth(task_id)
+71
View File
@@ -0,0 +1,71 @@
"""Error types for task plane operations."""
from __future__ import annotations
class TaskPlaneError(Exception):
"""Base exception for API-compatible failures."""
status_code = 500
error_code = "task_plane_error"
def __init__(self, message: str, *, code: str | None = None, details: dict | None = None):
super().__init__(message)
self.message = message
self.details = details or {}
if code:
self.error_code = code
def to_payload(self) -> dict:
return {"error": self.error_code, "message": self.message, "details": self.details}
class TaskPlaneHTTPError(TaskPlaneError):
"""Base class with explicit HTTP status."""
status_code = 500
class ValidationError(TaskPlaneHTTPError):
status_code = 400
error_code = "validation_error"
class UnauthorizedError(TaskPlaneHTTPError):
status_code = 401
error_code = "unauthorized"
class ForbiddenError(TaskPlaneHTTPError):
status_code = 403
error_code = "forbidden"
class NotFoundError(TaskPlaneHTTPError):
status_code = 404
error_code = "not_found"
class ConflictError(TaskPlaneHTTPError):
status_code = 409
error_code = "conflict"
class PayloadTooLargeError(TaskPlaneHTTPError):
status_code = 413
error_code = "payload_too_large"
class FailedDependencyError(TaskPlaneHTTPError):
status_code = 424
error_code = "failed_dependency"
class LockedError(TaskPlaneHTTPError):
status_code = 423
error_code = "locked"
class ServiceUnavailableError(TaskPlaneHTTPError):
status_code = 503
error_code = "service_unavailable"
+108
View File
@@ -0,0 +1,108 @@
"""CloudEvents-compatible event envelope and durable outbox helpers."""
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime, timezone
from typing import Callable, Protocol
from .store import TaskPlaneStore
@dataclass
class CloudEvent:
specversion: str = "1.0"
id: str = ""
type: str = ""
source: str = "spiffe://aios/task-plane"
subject: str = ""
datacontenttype: str = "application/json"
data: dict | None = None
@property
def time(self) -> str:
return datetime.now(timezone.utc).replace(microsecond=0).isoformat().replace("+00:00", "Z")
def asdict(self) -> dict:
return {
"specversion": self.specversion,
"id": self.id,
"type": self.type,
"source": self.source,
"subject": self.subject,
"time": self.time,
"datacontenttype": self.datacontenttype,
"data": self.data or {},
}
NOT_DELIVERED = "NOT_DELIVERED"
DELIVERED = "DELIVERED"
FAILED_DELIVERY = "FAILED"
class NotificationAdapter(Protocol):
name: str
def deliver(self, envelope: dict) -> str | None:
"""Return a provider receipt, or None when delivery is unconfirmed."""
class CallbackNotificationAdapter:
"""Adapter seam for webhook, NATS, GPT, dashboard, and WeChat transports."""
def __init__(self, name: str, callback: Callable[[dict], str | None]):
self.name = name
self._callback = callback
def deliver(self, envelope: dict) -> str | None:
return self._callback(envelope)
class WebhookAdapter(CallbackNotificationAdapter):
def __init__(self, callback: Callable[[dict], str | None]):
super().__init__("webhook", callback)
class NatsJetStreamAdapter(CallbackNotificationAdapter):
def __init__(self, callback: Callable[[dict], str | None]):
super().__init__("nats-jetstream", callback)
class GPTResultCallbackAdapter(CallbackNotificationAdapter):
def __init__(self, callback: Callable[[dict], str | None]):
super().__init__("gpt-result-callback", callback)
class DashboardAdapter(CallbackNotificationAdapter):
def __init__(self, callback: Callable[[dict], str | None]):
super().__init__("dashboard", callback)
class WeChatGatewayAdapter(CallbackNotificationAdapter):
def __init__(self, callback: Callable[[dict], str | None]):
super().__init__("wechat-gateway", callback)
class DurableOutboxDispatcher:
def __init__(self, store: TaskPlaneStore):
self.store = store
def dispatch(self, message_id: str, adapter: NotificationAdapter) -> dict:
row = self.store.get_outbox_message(message_id)
import json
envelope = json.loads(row["payload"])
try:
receipt = adapter.deliver(envelope)
except Exception:
return self.store.record_outbox_delivery(
message_id,
receipt=None,
error=f"{adapter.name} delivery failed",
)
return self.store.record_outbox_delivery(
message_id,
receipt=str(receipt) if receipt else None,
error=None if receipt else f"{adapter.name} receipt unavailable",
)
+31
View File
@@ -0,0 +1,31 @@
"""Evidence helpers."""
from __future__ import annotations
import hashlib
import json
from .store import TaskPlaneStore
def evidence_digest(payload: dict) -> str:
body = json.dumps(payload, sort_keys=True, separators=(",", ":"))
return hashlib.sha256(body.encode("utf-8")).hexdigest()
class EvidenceService:
def __init__(self, store: TaskPlaneStore):
self.store = store
def add(self, task_id: str, kind: str, payload: dict, worker_identity: str) -> str:
return self.store.add_evidence(task_id, kind, payload, worker_identity)
def list(self, task_id: str) -> list[dict]:
return self.store.list_evidence(task_id)
def verify(self, task_id: str, required_kinds: list[str]) -> bool:
evidences = self.list(task_id)
observed = {e["kind"] for e in evidences}
return set(required_kinds).issubset(
observed
) and self.store.verify_evidence_integrity(task_id)
+22
View File
@@ -0,0 +1,22 @@
"""Event journaling helpers."""
from __future__ import annotations
from .errors import ValidationError
from .store import TaskPlaneStore
class TaskJournal:
def __init__(self, store: TaskPlaneStore):
self.store = store
def append(self, task_id: str, event_type: str, data: dict, *, subject: str | None = None) -> None:
if not event_type:
raise ValidationError("event_type required")
self.store.append_event(task_id, event_type, data, subject=subject)
def list(self, task_id: str) -> list[dict]:
return self.store.list_events(task_id)
def check_chain(self, task_id: str) -> bool:
return self.store.verify_event_chain(task_id)
+32
View File
@@ -0,0 +1,32 @@
"""Claim and lease management helpers."""
from __future__ import annotations
from .errors import LockedError
from .models import ProbeEvidence
from .store import TaskPlaneStore
class LeaseManager:
def __init__(self, store: TaskPlaneStore):
self.store = store
def claim(self, task_id: str, worker_id: str, worker_identity: str, lease_seconds: int, expected_revision: int, operation_digest: str):
return self.store.claim_task(task_id, worker_id, worker_identity, lease_seconds, expected_revision, operation_digest)
def heartbeat(self, task_id: str, claim_id: str, lease_seconds: int):
claim = self.store.heartbeat_claim(task_id, claim_id, lease_seconds)
return claim
def probe(self, claim_id: str):
claim = self.store.get_claim(claim_id)
if claim.status != "ACTIVE":
raise LockedError("claim not active")
return claim
def recover(self, task_id: str, claim_id: str, probe: ProbeEvidence):
return self.store.recover_expired_claim(
task_id,
claim_id,
probe,
)
+829
View File
@@ -0,0 +1,829 @@
"""Core data models for the task plane."""
from __future__ import annotations
import hashlib
import json
import re
import time
import uuid
import urllib.parse
from enum import StrEnum
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any
SENSITIVE_KEYS = {
"secret",
"password",
"token",
"api_key",
"access_token",
"refresh_token",
"authorization",
"private_key",
"client_secret",
"credential",
}
IDENTIFIER_RE = re.compile(r"[A-Za-z0-9][A-Za-z0-9._:-]{0,127}")
TYPE_RE = re.compile(r"[A-Za-z0-9][A-Za-z0-9._-]{0,127}")
RISK_LEVELS = {"low", "medium", "high"}
DATA_CLASSIFICATIONS = {
"public",
"internal",
"confidential",
"credential-secret",
}
SECRET_TEXT_PATTERNS = (
re.compile(r"(?i)\bAuthorization\s*:\s*Bearer\s+\S+"),
re.compile(r"(?i)\b(?:password|passwd|api[_-]?key|access[_-]?token|secret)\s*[:=]\s*\S+"),
re.compile(r"\bsk-[A-Za-z0-9_-]{12,}\b"),
re.compile(r"\bAKIA[0-9A-Z]{16}\b"),
re.compile(r"-----BEGIN (?:RSA |EC |OPENSSH )?PRIVATE KEY-----"),
)
TASK_PLAN_KEYS = {
"schema_version",
"task_id",
"task_type",
"issuer",
"intent",
"required_capabilities",
"preferred_workers",
"risk",
"approval_policy",
"dependencies",
"max_delegation_depth",
"budgets",
"evidence_requirements",
"parent_task_id",
"correlation_id",
"metadata",
"data_classification",
"execution_zone",
"network_policy",
"logging_policy",
"allowed_input_paths",
"allowed_output_paths",
"output_policy",
}
def _now_ts() -> float:
return time.time()
def canonical_json(value: Any) -> str:
return json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
def compute_sha256_hex(value: Any) -> str:
return hashlib.sha256(canonical_json(value).encode("utf-8")).hexdigest()
def contains_secret_text(value: Any) -> bool:
if isinstance(value, str):
return any(pattern.search(value) for pattern in SECRET_TEXT_PATTERNS)
if isinstance(value, dict):
return any(contains_secret_text(item) for item in value.values())
if isinstance(value, (list, tuple)):
return any(contains_secret_text(item) for item in value)
return False
def validate_relative_paths(values: Any, name: str) -> list[str]:
paths = _normalize_list(values)
for value in paths:
decoded = urllib.parse.unquote(value)
parts = value.replace("\\", "/").split("/")
if (
decoded != value
or "\\" in value
or value.startswith("/")
or re.match(r"^[A-Za-z]:", value)
or any(part in {"", ".", ".."} for part in parts)
):
raise ValueError(f"invalid {name}")
return paths
def validate_identifier(value: Any, name: str) -> str:
text = str(value or "")
if (
not IDENTIFIER_RE.fullmatch(text)
or urllib.parse.unquote(text) != text
or any(ord(ch) < 32 or ord(ch) == 127 for ch in text)
):
raise ValueError(f"invalid {name}")
return text
def validate_type(value: Any, name: str) -> str:
text = str(value or "")
if not TYPE_RE.fullmatch(text):
raise ValueError(f"invalid {name}")
return text
def redact_value(value: Any) -> Any:
if isinstance(value, dict):
redacted = {}
for key, inner in value.items():
lowered = str(key).lower()
if lowered in SENSITIVE_KEYS or lowered.endswith(
("_password", "_secret", "_access_token", "_refresh_token", "_private_key")
):
redacted[key] = "[REDACTED]"
else:
redacted[key] = redact_value(inner)
return redacted
if isinstance(value, list):
return [redact_value(v) for v in value]
if isinstance(value, tuple):
return tuple(redact_value(v) for v in value)
return value
class TaskStateTransitionError(ValueError):
pass
class TaskStatus(StrEnum):
CREATED = "CREATED"
QUEUED = "QUEUED"
CLAIMED = "CLAIMED"
RUNNING = "RUNNING"
WAITING_CHILD = "WAITING_CHILD"
VERIFYING = "VERIFYING"
COMPLETED = "COMPLETED"
FAILED_RECOVERABLE = "FAILED_RECOVERABLE"
FAILED_FINAL = "FAILED_FINAL"
CANCELLED = "CANCELLED"
MANUAL_REVIEW_REQUIRED = "MANUAL_REVIEW_REQUIRED"
TASK_STATES = {
status.value for status in TaskStatus
}
TASK_NEXT_STATES = {
"CREATED": {"QUEUED", "CLAIMED", "CANCELLED", "MANUAL_REVIEW_REQUIRED"},
"QUEUED": {"CLAIMED", "CANCELLED", "FAILED_RECOVERABLE", "FAILED_FINAL"},
"CLAIMED": {"RUNNING", "FAILED_RECOVERABLE", "FAILED_FINAL", "CANCELLED"},
"RUNNING": {"VERIFYING", "WAITING_CHILD", "COMPLETED", "FAILED_RECOVERABLE", "FAILED_FINAL", "CANCELLED"},
"WAITING_CHILD": {"RUNNING", "VERIFYING", "FAILED_RECOVERABLE", "FAILED_FINAL", "CANCELLED"},
"VERIFYING": {"COMPLETED", "FAILED_RECOVERABLE", "FAILED_FINAL", "CANCELLED"},
"COMPLETED": set(),
"FAILED_RECOVERABLE": {"CLAIMED", "RUNNING", "CANCELLED"},
"FAILED_FINAL": set(),
"CANCELLED": set(),
"MANUAL_REVIEW_REQUIRED": {"RUNNING", "CANCELLED"},
}
@dataclass
class Budget:
time_seconds: int | None = None
attempts: int | None = None
tokens: int | None = None
cost: float | None = None
@classmethod
def from_dict(cls, data: dict[str, Any] | None) -> "Budget":
if not data:
return cls()
if not isinstance(data, dict) or set(data) - {
"time_seconds",
"attempts",
"tokens",
"cost",
}:
raise ValueError("invalid budget")
for name in ("time_seconds", "attempts", "tokens", "cost"):
value = data.get(name)
if value is not None:
try:
if float(value) < 0:
raise ValueError("budget values must be non-negative")
except (TypeError, ValueError) as exc:
raise ValueError("invalid budget value") from exc
return cls(
time_seconds=_to_positive_int(data.get("time_seconds")),
attempts=_to_positive_int(data.get("attempts")),
tokens=_to_positive_int(data.get("tokens")),
cost=_to_float(data.get("cost")),
)
def to_dict(self) -> dict[str, Any]:
return {
"time_seconds": self.time_seconds,
"attempts": self.attempts,
"tokens": self.tokens,
"cost": self.cost,
}
def merge(self, other: "Budget") -> "Budget":
return Budget(
time_seconds=_first_non_none(self.time_seconds, other.time_seconds),
attempts=_first_non_none(self.attempts, other.attempts),
tokens=_first_non_none(self.tokens, other.tokens),
cost=_first_non_none(self.cost, other.cost),
)
@dataclass
class TaskPlan:
schema_version: int = 1
task_id: str | None = None
task_type: str = "code.change"
issuer: dict[str, Any] = field(default_factory=dict)
intent: str | None = None
required_capabilities: list[str] = field(default_factory=list)
preferred_workers: list[str] = field(default_factory=list)
risk: str = "medium"
approval_policy: str = "patch-only"
dependencies: list[str] = field(default_factory=list)
max_delegation_depth: int = 5
budgets: Budget = field(default_factory=Budget)
evidence_requirements: list[str] = field(default_factory=list)
parent_task_id: str | None = None
correlation_id: str | None = None
metadata: dict[str, Any] = field(default_factory=dict)
data_classification: str = "internal"
execution_zone: str = "isolated-development"
network_policy: str = "controlled"
logging_policy: str = "redacted"
allowed_input_paths: list[str] = field(default_factory=list)
allowed_output_paths: list[str] = field(default_factory=list)
output_policy: str = "standard"
@classmethod
def from_dict(cls, data: dict[str, Any] | None) -> "TaskPlan":
if not isinstance(data, dict):
raise ValueError("task plan payload must be an object")
unknown = set(data) - TASK_PLAN_KEYS
if unknown:
raise ValueError("unknown task plan fields")
if "task_id" not in data or data["task_id"] is None:
data = dict(data)
data["task_id"] = f"task-{uuid.uuid4()}"
risk = str(data.get("risk", "medium"))
if risk not in RISK_LEVELS:
raise ValueError("invalid risk")
issuer = data.get("issuer", {})
if not isinstance(issuer, dict):
raise ValueError("issuer must be an object")
schema_version = int(data.get("schema_version", 1))
if schema_version != 1:
raise ValueError("unsupported schema_version")
max_depth = _to_positive_int(data.get("max_delegation_depth"), default=5)
if max_depth is None or max_depth > 32:
raise ValueError("invalid max_delegation_depth")
classification = str(data.get("data_classification", "internal"))
if classification not in DATA_CLASSIFICATIONS:
raise ValueError("invalid data_classification")
execution_zone = validate_type(
data.get("execution_zone", "isolated-development"), "execution_zone"
)
network_policy = validate_type(
data.get("network_policy", "controlled"), "network_policy"
)
logging_policy = validate_type(
data.get("logging_policy", "redacted"), "logging_policy"
)
output_policy = validate_type(
data.get("output_policy", "standard"), "output_policy"
)
if contains_secret_text(
{
"intent": data.get("intent"),
"metadata": data.get("metadata", {}),
}
):
raise ValueError("task content contains secret material")
if classification == "credential-secret":
if execution_zone != "secure-file-transformer":
raise ValueError("credential-secret requires secure-file-transformer")
if network_policy != "denied":
raise ValueError("credential-secret requires network denied")
if logging_policy != "metadata-only":
raise ValueError("credential-secret requires metadata-only logging")
if output_policy != "hash-count-status-only":
raise ValueError("credential-secret requires restricted output")
return cls(
schema_version=schema_version,
task_id=validate_identifier(data["task_id"], "task_id"),
task_type=validate_type(data.get("task_type", "code.change"), "task_type"),
issuer=issuer,
intent=data.get("intent"),
required_capabilities=_normalize_list(data.get("required_capabilities", [])),
preferred_workers=_normalize_list(data.get("preferred_workers", [])),
risk=risk,
approval_policy=validate_type(
data.get("approval_policy", "patch-only"), "approval_policy"
),
dependencies=_normalize_list(data.get("dependencies", [])),
max_delegation_depth=max_depth,
budgets=Budget.from_dict(data.get("budgets")),
evidence_requirements=_normalize_list(data.get("evidence_requirements", [])),
parent_task_id=data.get("parent_task_id"),
correlation_id=data.get("correlation_id"),
metadata=data.get("metadata", {}) if isinstance(data.get("metadata", {}), dict) else {},
data_classification=classification,
execution_zone=execution_zone,
network_policy=network_policy,
logging_policy=logging_policy,
allowed_input_paths=validate_relative_paths(
data.get("allowed_input_paths", []), "allowed_input_paths"
),
allowed_output_paths=validate_relative_paths(
data.get("allowed_output_paths", []), "allowed_output_paths"
),
output_policy=output_policy,
)
def to_dict(self, *, redact: bool = False) -> dict[str, Any]:
payload = {
"schema_version": self.schema_version,
"task_id": self.task_id,
"task_type": self.task_type,
"issuer": self.issuer,
"intent": self.intent,
"required_capabilities": list(self.required_capabilities),
"preferred_workers": list(self.preferred_workers),
"risk": self.risk,
"approval_policy": self.approval_policy,
"dependencies": list(self.dependencies),
"max_delegation_depth": self.max_delegation_depth,
"budgets": self.budgets.to_dict(),
"evidence_requirements": list(self.evidence_requirements),
"parent_task_id": self.parent_task_id,
"correlation_id": self.correlation_id,
"metadata": dict(self.metadata),
"data_classification": self.data_classification,
"execution_zone": self.execution_zone,
"network_policy": self.network_policy,
"logging_policy": self.logging_policy,
"allowed_input_paths": list(self.allowed_input_paths),
"allowed_output_paths": list(self.allowed_output_paths),
"output_policy": self.output_policy,
}
if redact:
payload = redact_value(payload)
return payload
def operation_digest(self) -> str:
return compute_sha256_hex(self.to_dict())
@dataclass
class TaskRecord:
task_id: str
status: str
revision: int
task_plan: TaskPlan
created_at: float
updated_at: float
root_task_id: str
idempotency_key: str | None = None
failure_reason: str | None = None
def with_status(self, status: str) -> "TaskRecord":
allowed = TASK_NEXT_STATES.get(self.status, set())
if status != self.status and (not allowed and self.status not in {"COMPLETED", "FAILED_FINAL", "CANCELLED"}):
raise TaskStateTransitionError(f"Cannot transition {self.status} -> {status}")
if status not in TASK_STATES:
raise TaskStateTransitionError(f"Invalid status {status}")
if status != self.status and self.status in TASK_NEXT_STATES and status not in TASK_NEXT_STATES[self.status]:
raise TaskStateTransitionError(f"Cannot transition {self.status} -> {status}")
return TaskRecord(
task_id=self.task_id,
status=status,
revision=self.revision + 1,
task_plan=self.task_plan,
created_at=self.created_at,
updated_at=_now_ts(),
root_task_id=self.root_task_id,
idempotency_key=self.idempotency_key,
failure_reason=self.failure_reason,
)
def to_dict(self, *, include_task=True, redact_payload: bool = True) -> dict[str, Any]:
payload = {
"task_id": self.task_id,
"status": self.status,
"revision": self.revision,
"root_task_id": self.root_task_id,
"parent_task_id": self.task_plan.parent_task_id,
"created_at": _fmt_ts(self.created_at),
"updated_at": _fmt_ts(self.updated_at),
"failure_reason": self.failure_reason,
}
if include_task:
payload["plan"] = self.task_plan.to_dict(redact=redact_payload)
if self.idempotency_key:
payload["idempotency_key"] = self.idempotency_key
return payload
@dataclass
class WorkerRegistration:
worker_id: str
worker_type: str
identity: str
capabilities: list[str]
trust_zone: str
max_concurrency: int
heartbeat_interval_seconds: int
supported_protocols: list[str]
registered_at: float
last_heartbeat: float
status: str = "active"
disabled: bool = False
owner_identity: str = ""
registration_revision: int = 1
registration_policy: str = "self"
created_at: float = 0.0
updated_at: float = 0.0
disabled_reason: str | None = None
@classmethod
def from_dict(cls, data: dict[str, Any]) -> "WorkerRegistration":
if not isinstance(data, dict):
raise ValueError("worker registration payload must be an object")
now = _now_ts()
worker_id = validate_identifier(data.get("worker_id"), "worker_id")
worker_type = validate_type(data.get("worker_type"), "worker_type")
identity = str(data.get("identity") or "").strip()
if not identity or any(ord(ch) < 32 for ch in identity):
raise ValueError("invalid identity")
max_concurrency = _to_positive_int(data.get("max_concurrency"), default=1)
heartbeat_interval = _to_positive_int(
data.get("heartbeat_interval_seconds"), default=15
)
if not max_concurrency:
raise ValueError("max_concurrency must be positive")
if not heartbeat_interval:
raise ValueError("heartbeat_interval_seconds must be positive")
return cls(
worker_id=worker_id,
worker_type=worker_type,
identity=identity,
capabilities=_normalize_list(data.get("capabilities", [])),
trust_zone=str(data.get("trust_zone") or "default"),
max_concurrency=max_concurrency,
heartbeat_interval_seconds=heartbeat_interval,
supported_protocols=_normalize_list(
data.get("supported_protocols", ["aios-task-v1"])
),
registered_at=float(data.get("registered_at") or now),
last_heartbeat=float(data.get("last_heartbeat") or now),
status=str(data.get("status") or "active"),
disabled=bool(data.get("disabled", False)),
owner_identity=str(data.get("owner_identity") or identity),
registration_revision=int(data.get("registration_revision") or 1),
registration_policy=str(data.get("registration_policy") or "self"),
created_at=float(data.get("created_at") or data.get("registered_at") or now),
updated_at=float(data.get("updated_at") or now),
disabled_reason=data.get("disabled_reason"),
)
def to_dict(self, redact_payload: bool = True) -> dict[str, Any]:
payload = {
"worker_id": self.worker_id,
"worker_type": self.worker_type,
"identity": self.identity,
"capabilities": list(self.capabilities),
"trust_zone": self.trust_zone,
"max_concurrency": self.max_concurrency,
"heartbeat_interval_seconds": self.heartbeat_interval_seconds,
"supported_protocols": list(self.supported_protocols),
"registered_at": _fmt_ts(self.registered_at),
"last_heartbeat": _fmt_ts(self.last_heartbeat),
"status": self.status,
"disabled": self.disabled,
"owner_identity": self.owner_identity,
"registration_revision": self.registration_revision,
"registration_policy": self.registration_policy,
"created_at": _fmt_ts(self.created_at or self.registered_at),
"updated_at": _fmt_ts(self.updated_at or self.last_heartbeat),
"disabled_reason": self.disabled_reason,
}
if redact_payload:
payload = redact_value(payload)
return payload
@dataclass(frozen=True)
class EvidenceEnvelope:
schema_version: int
evidence_id: str
task_id: str
claim_id: str
kind: str
worker_identity: str
payload: dict[str, Any]
created_at: float
previous_evidence_hash: str
evidence_digest: str
signature: str | None = None
attestation_ref: str | None = None
@classmethod
def create(
cls,
*,
task_id: str,
claim_id: str,
kind: str,
worker_identity: str,
payload: dict[str, Any],
previous_evidence_hash: str,
signature: str | None = None,
attestation_ref: str | None = None,
evidence_id: str | None = None,
created_at: float | None = None,
) -> "EvidenceEnvelope":
envelope = cls(
schema_version=1,
evidence_id=evidence_id or f"evd-{uuid.uuid4()}",
task_id=validate_identifier(task_id, "task_id"),
claim_id=validate_identifier(claim_id, "claim_id"),
kind=validate_type(kind, "evidence kind"),
worker_identity=str(worker_identity),
payload=redact_value(payload),
created_at=float(created_at or _now_ts()),
previous_evidence_hash=str(previous_evidence_hash or ""),
evidence_digest="",
signature=signature,
attestation_ref=attestation_ref,
)
return dataclass_replace(
envelope, evidence_digest=compute_sha256_hex(envelope.digest_payload())
)
def digest_payload(self) -> dict[str, Any]:
return {
"schema_version": self.schema_version,
"evidence_id": self.evidence_id,
"task_id": self.task_id,
"claim_id": self.claim_id,
"kind": self.kind,
"worker_identity": self.worker_identity,
"payload": self.payload,
"created_at": self.created_at,
"previous_evidence_hash": self.previous_evidence_hash,
"attestation_ref": self.attestation_ref,
}
def verify(
self,
*,
expected_task_id: str,
expected_claim_id: str,
expected_previous_hash: str,
attestation_verifier: Any = None,
) -> bool:
if (
self.schema_version != 1
or self.task_id != expected_task_id
or self.claim_id != expected_claim_id
or self.previous_evidence_hash != expected_previous_hash
or self.evidence_digest != compute_sha256_hex(self.digest_payload())
):
return False
if self.signature or self.attestation_ref:
return True if attestation_verifier is None else bool(attestation_verifier(self))
return True
def to_dict(self) -> dict[str, Any]:
return {
**self.digest_payload(),
"evidence_digest": self.evidence_digest,
"signature": self.signature,
}
class ProbeOutcome(StrEnum):
NOT_APPLIED = "NOT_APPLIED"
APPLIED_VERIFIED = "APPLIED_VERIFIED"
INDETERMINATE = "INDETERMINATE"
CONFLICT = "CONFLICT"
@dataclass(frozen=True)
class ProbeEvidence:
schema_version: int
probe_id: str
task_id: str
claim_id: str
worker_id: str
worker_identity: str
operation_digest: str
outcome: ProbeOutcome
observed_at: float
expires_at: float
probe_executor_identity: str
evidence_digest: str
signature: str | None = None
attestation_ref: str | None = None
@classmethod
def create(cls, **data: Any) -> "ProbeEvidence":
outcome = data["outcome"]
if not isinstance(outcome, ProbeOutcome):
outcome = ProbeOutcome(str(outcome))
probe = cls(
schema_version=1,
probe_id=data.get("probe_id") or f"probe-{uuid.uuid4()}",
task_id=validate_identifier(data["task_id"], "task_id"),
claim_id=validate_identifier(data["claim_id"], "claim_id"),
worker_id=validate_identifier(data["worker_id"], "worker_id"),
worker_identity=str(data["worker_identity"]),
operation_digest=str(data["operation_digest"]),
outcome=outcome,
observed_at=float(data["observed_at"]),
expires_at=float(data["expires_at"]),
probe_executor_identity=str(data["probe_executor_identity"]),
evidence_digest="",
signature=data.get("signature"),
attestation_ref=data.get("attestation_ref"),
)
return dataclass_replace(
probe, evidence_digest=compute_sha256_hex(probe.digest_payload())
)
def digest_payload(self) -> dict[str, Any]:
return {
"schema_version": self.schema_version,
"probe_id": self.probe_id,
"task_id": self.task_id,
"claim_id": self.claim_id,
"worker_id": self.worker_id,
"worker_identity": self.worker_identity,
"operation_digest": self.operation_digest,
"outcome": self.outcome.value,
"observed_at": self.observed_at,
"expires_at": self.expires_at,
"probe_executor_identity": self.probe_executor_identity,
"attestation_ref": self.attestation_ref,
}
def verify(self, *, now: float, attestation_verifier: Any) -> bool:
if (
self.schema_version != 1
or self.expires_at <= now
or self.observed_at > now
or self.evidence_digest != compute_sha256_hex(self.digest_payload())
or not (self.signature or self.attestation_ref)
):
return False
return bool(attestation_verifier and attestation_verifier(self))
def to_dict(self) -> dict[str, Any]:
return {
**self.digest_payload(),
"evidence_digest": self.evidence_digest,
"signature": self.signature,
}
def dataclass_replace(value: Any, **changes: Any) -> Any:
from dataclasses import replace
return replace(value, **changes)
@dataclass
class WorkerClaim:
claim_id: str
task_id: str
worker_id: str
worker_identity: str
claimed_at: float
lease_until: float
attempt: int
task_revision: int
operation_digest: str
status: str = "ACTIVE"
def to_dict(self, redact_payload: bool = True) -> dict[str, Any]:
payload = {
"claim_id": self.claim_id,
"task_id": self.task_id,
"worker_id": self.worker_id,
"worker_identity": self.worker_identity,
"claimed_at": _fmt_ts(self.claimed_at),
"lease_until": _fmt_ts(self.lease_until),
"attempt": self.attempt,
"task_revision": self.task_revision,
"operation_digest": self.operation_digest,
"status": self.status,
}
if redact_payload:
payload = redact_value(payload)
return payload
@dataclass
class ResultEnvelope:
task_id: str
claim_id: str
worker_identity: str
status: str
outputs: list[dict[str, Any]]
evidence: list[dict[str, Any]]
started_at: str
completed_at: str
result_digest: str | None = None
signature: str | None = None
attestation_ref: str | None = None
operation_digest: str | None = None
trust_status: str = "untrusted_shadow_result"
def digest_payload(self) -> dict[str, Any]:
return {
"task_id": self.task_id,
"claim_id": self.claim_id,
"worker_identity": self.worker_identity,
"status": self.status,
"outputs": self.outputs,
"evidence": self.evidence,
"started_at": self.started_at,
"completed_at": self.completed_at,
"operation_digest": self.operation_digest,
"attestation_ref": self.attestation_ref,
}
def computed_digest(self) -> str:
return compute_sha256_hex(self.digest_payload())
def to_dict(self, redact_payload: bool = True) -> dict[str, Any]:
payload = {
"task_id": self.task_id,
"claim_id": self.claim_id,
"worker_identity": self.worker_identity,
"status": self.status,
"outputs": self.outputs,
"evidence": self.evidence,
"started_at": self.started_at,
"completed_at": self.completed_at,
"result_digest": self.result_digest,
"signature": self.signature,
"attestation_ref": self.attestation_ref,
"operation_digest": self.operation_digest,
"trust_status": self.trust_status,
}
if redact_payload:
payload = redact_value(payload)
return payload
def _normalize_list(value: Any) -> list[str]:
if not isinstance(value, list):
raise ValueError("expected list")
if len(value) > 128 or any(
not isinstance(v, str)
or not v
or len(v) > 256
or any(ord(ch) < 32 for ch in v)
for v in value
):
raise ValueError("invalid list item")
return list(value)
def _to_positive_int(value: Any, default: int | None = None) -> int | None:
if value is None:
return default
try:
number = int(value)
except (TypeError, ValueError):
return default
return number if number >= 0 else default
def _to_float(value: Any) -> float | None:
if value is None:
return None
try:
return float(value)
except (TypeError, ValueError):
return None
def _first_non_none(primary: Any, fallback: Any) -> Any:
if primary is not None:
return primary
return fallback
def iso8601_now() -> str:
return datetime.now(timezone.utc).replace(microsecond=0).isoformat().replace("+00:00", "Z")
def _fmt_ts(value: float | None) -> str | None:
if value is None:
return None
return datetime.fromtimestamp(value, tz=timezone.utc).replace(microsecond=0).isoformat().replace("+00:00", "Z")
+226
View File
@@ -0,0 +1,226 @@
"""Policy and verification logic (fail-closed)."""
from __future__ import annotations
from dataclasses import dataclass
from typing import Callable
from .errors import ForbiddenError, LockedError
from .models import ProbeEvidence, ResultEnvelope, TaskRecord
RISK_RANK = {"low": 1, "medium": 2, "high": 3}
READONLY_WORKER_ACTIONS = frozenset(
{"claim", "result", "fail", "progress", "heartbeat"}
)
ALL_ACTION_CAPABILITIES = frozenset(
{
"task.creator",
"worker.self_register",
"worker.admin",
"task.claim",
"task.cancel",
"task.delegate",
"task.read",
"task.audit",
}
)
@dataclass
class PolicyDecision:
allowed: bool
reason: str = ""
@dataclass(frozen=True)
class ReadonlyWorkerProfile:
"""A narrowly-scoped identity allowed to process safe Shadow tasks only."""
worker_id: str
identity: str
capabilities: frozenset[str]
task_types: frozenset[str] = frozenset({"code.review", "test.execute"})
def permits_task(self, task: TaskRecord) -> bool:
plan = task.task_plan
if plan.task_type not in self.task_types or plan.risk != "low":
return False
if plan.data_classification not in {"public", "internal"}:
return False
if plan.metadata.get("production_side_effects") is not False:
return False
return set(plan.required_capabilities).issubset(self.capabilities)
class IdentityVerifier:
def __init__(self, trusted_identities: set[str] | None = None):
self.trusted_identities = trusted_identities
def verify(self, identity: str) -> bool:
if self.trusted_identities is None:
return True
return identity in self.trusted_identities
class EvidenceVerifier:
def __init__(self, mandatory: list[str] | None = None):
self.mandatory = set(mandatory or [])
def verify(self, task: TaskRecord, evidences: list[dict]) -> PolicyDecision:
required = set(task.task_plan.evidence_requirements or [])
if not required and not self.mandatory:
return PolicyDecision(True)
observed = {e.get("kind") for e in evidences}
missing = required.union(self.mandatory) - observed
if missing:
return PolicyDecision(False, f"missing evidence: {', '.join(sorted(missing))}")
return PolicyDecision(True)
class PolicyEngine:
"""
Fail-closed policy engine with minimal but explicit decisions.
"""
def __init__(
self,
*,
identity_verifier: IdentityVerifier | None = None,
evidence_verifier: EvidenceVerifier | None = None,
execution_enabled: bool = False,
max_risk: str = "medium",
result_signature_verifier: Callable[[ResultEnvelope], bool] | None = None,
probe_attestation_verifier: Callable[[ProbeEvidence], bool] | None = None,
identity_capabilities: dict[str, set[str]] | None = None,
admin_identities: set[str] | None = None,
readonly_worker_execution_enabled: bool = False,
readonly_worker_profiles: list[ReadonlyWorkerProfile] | None = None,
) -> None:
if max_risk not in RISK_RANK:
raise ValueError("invalid max_risk")
self.identity_verifier = identity_verifier or IdentityVerifier()
self.evidence_verifier = evidence_verifier or EvidenceVerifier()
self.execution_enabled = execution_enabled
self.max_risk = max_risk
self.result_signature_verifier = result_signature_verifier
self.probe_attestation_verifier = probe_attestation_verifier
self.identity_capabilities = {
identity: set(capabilities)
for identity, capabilities in (identity_capabilities or {}).items()
}
self.admin_identities = set(admin_identities or ())
self.readonly_worker_execution_enabled = readonly_worker_execution_enabled
self.readonly_worker_profiles = {
profile.identity: profile for profile in (readonly_worker_profiles or [])
}
def _readonly_worker_allows(
self, action: str, actor: str, task: TaskRecord | None
) -> bool:
profile = self.readonly_worker_profiles.get(actor)
return bool(
self.readonly_worker_execution_enabled
and action in READONLY_WORKER_ACTIONS
and profile is not None
and (task is None or profile.permits_task(task))
)
def assert_readonly_worker_binding(self, actor: str, worker_id: str) -> None:
profile = self.readonly_worker_profiles.get(actor)
if profile is not None and profile.worker_id != worker_id:
raise ForbiddenError("worker identity binding mismatch")
def assert_readonly_worker_registration(self, actor: str, payload: dict) -> None:
profile = self.readonly_worker_profiles.get(actor)
if profile is None:
return
if (
str(payload.get("worker_id") or "") != profile.worker_id
or str(payload.get("identity") or "") != profile.identity
or set(payload.get("capabilities") or []) != set(profile.capabilities)
or str(payload.get("trust_zone") or "") != "isolated-review"
or int(payload.get("max_concurrency") or 0) != 1
or int(payload.get("heartbeat_interval_seconds") or 0) != 15
):
raise ForbiddenError("readonly worker registration mismatch")
def verify_result(
self, result: ResultEnvelope, task: TaskRecord | None = None
) -> PolicyDecision:
computed = result.computed_digest()
if result.result_digest and result.result_digest != computed:
return PolicyDecision(False, "result digest mismatch")
sensitive = bool(
task
and (
task.task_plan.data_classification
in {"confidential", "credential-secret"}
or any(
capability.startswith(
("code.write", "deploy", "secret", "production")
)
for capability in task.task_plan.required_capabilities
)
)
)
if sensitive and not (result.signature or result.attestation_ref):
return PolicyDecision(False, "trusted result proof required")
if result.signature is not None or result.attestation_ref is not None:
if self.result_signature_verifier is None:
return PolicyDecision(False, "result signature verifier unavailable")
if not self.result_signature_verifier(result):
return PolicyDecision(False, "result signature invalid")
result.trust_status = "trusted_result"
else:
result.trust_status = "untrusted_shadow_result"
return PolicyDecision(True)
def authorize(self, action: str, actor: str, task: TaskRecord | None = None, context: dict | None = None) -> PolicyDecision:
context = context or {}
if not self.identity_verifier.verify(actor):
return PolicyDecision(False, "identity untrusted")
action_capabilities = {
"create": "task.creator",
"read": "task.read",
"audit": "task.audit",
"claim": "task.claim",
"result": "task.claim",
"fail": "task.claim",
"progress": "task.claim",
"heartbeat": "task.claim",
"cancel": "task.cancel",
"delegate": "task.delegate",
"worker.register": "worker.self_register",
"worker.admin": "worker.admin",
}
required_capability = action_capabilities.get(action)
if required_capability:
granted = set(
self.identity_capabilities.get(
actor, self.identity_capabilities.get("*", set())
)
)
if required_capability not in granted:
return PolicyDecision(False, f"action {action} not permitted")
if action in {"claim", "result", "fail", "cancel", "progress", "delegate", "heartbeat"} and not self.execution_enabled:
if not self._readonly_worker_allows(action, actor, task):
return PolicyDecision(False, "execution disabled")
if action == "delegate" and task and task.status not in {"COMPLETED", "FAILED_FINAL", "CANCELLED"}:
risk = str(context.get("risk", task.task_plan.risk))
if RISK_RANK.get(risk, 2) > RISK_RANK.get(self.max_risk, 1):
return PolicyDecision(False, f"risk {risk} exceeds policy")
return PolicyDecision(True)
def assert_allowed(self, action: str, actor: str, task: TaskRecord | None = None, context: dict | None = None):
decision = self.authorize(action, actor, task, context=context)
if not decision.allowed:
if decision.reason == "identity untrusted":
raise ForbiddenError(decision.reason)
if decision.reason == "execution disabled":
raise LockedError("execution disabled")
raise ForbiddenError(decision.reason)
@@ -0,0 +1,76 @@
"""Task scheduler primitives."""
from __future__ import annotations
import time
from .errors import NotFoundError
from .capabilities import has_capabilities
from .models import WorkerClaim
from .store import TaskPlaneStore
class TaskScheduler:
def __init__(self, store: TaskPlaneStore):
self.store = store
def next_queued_task(self, worker_id: str):
worker = self.store.get_worker(worker_id)
tasks = sorted(
self.store.list_tasks(),
key=lambda task: (
-int(task.task_plan.metadata.get("priority", 0)),
task.created_at,
task.task_id,
),
)
for task in tasks:
if task.status not in {"CREATED", "QUEUED", "FAILED_RECOVERABLE"}:
continue
if task.task_plan.task_type == "worker.registry.audit":
continue
if task.status == "FAILED_RECOVERABLE" and not self.store.task_has_retryable_probe(
task.task_id
):
continue
if not has_capabilities(
worker.capabilities, task.task_plan.required_capabilities
):
continue
if any(
self.store.get_task(dependency).status != "COMPLETED"
for dependency in task.task_plan.dependencies
):
continue
if (
task.task_plan.data_classification == "credential-secret"
and worker.worker_type != "secure_file_transformer"
and "secure_file_transformer" not in worker.capabilities
):
continue
return task
return None
def schedule_for_worker(self, worker_id: str, lease_seconds: int = 15) -> WorkerClaim:
worker = self.store.get_worker(worker_id)
if worker.disabled:
from .errors import LockedError
raise LockedError("worker disabled")
if time.time() > worker.last_heartbeat + (
worker.heartbeat_interval_seconds * 3
):
from .errors import LockedError
raise LockedError("worker offline")
task = self.next_queued_task(worker_id)
if task is None:
raise NotFoundError("no queued tasks available")
claim = self.store.claim_task(
task.task_id,
worker_id,
worker.identity,
lease_seconds=lease_seconds,
expected_revision=task.revision,
operation_digest=task.task_plan.operation_digest(),
)
return claim
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,68 @@
"""Worker registry abstraction and read helpers."""
from __future__ import annotations
import time
from .capabilities import has_capabilities
from .errors import NotFoundError
from .models import WorkerRegistration
from .store import TaskPlaneStore
class WorkerRegistry:
def __init__(self, store: TaskPlaneStore):
self.store = store
def register(
self,
payload: dict,
*,
actor_identity: str,
is_admin: bool,
expected_revision: int | None,
) -> WorkerRegistration:
reg = WorkerRegistration.from_dict(payload)
return self.store.register_worker(
reg,
actor_identity=actor_identity,
is_admin=is_admin,
expected_revision=expected_revision,
)
def heartbeat(
self,
worker_id: str,
heartbeat_interval_seconds: int | None = None,
*,
actor_identity: str,
expected_revision: int | None,
) -> WorkerRegistration:
return self.store.heartbeat_worker(
worker_id,
heartbeat_interval_seconds=heartbeat_interval_seconds,
actor_identity=actor_identity,
expected_revision=expected_revision,
)
def list(self):
return self.store.list_workers()
def get(self, worker_id: str) -> WorkerRegistration:
try:
return self.store.get_worker(worker_id)
except NotFoundError:
raise
def can_execute(self, worker_id: str, required_capabilities: list[str]) -> bool:
worker = self.store.get_worker(worker_id)
return has_capabilities(worker.capabilities, required_capabilities)
def is_online(self, worker_id: str, *, now: float | None = None) -> bool:
worker = self.store.get_worker(worker_id)
deadline = worker.last_heartbeat + (worker.heartbeat_interval_seconds * 3)
return not worker.disabled and (time.time() if now is None else now) <= deadline
def stale_workers(self, *, now: float | None = None) -> list[WorkerRegistration]:
current = time.time() if now is None else now
return [worker for worker in self.list() if not self.is_online(worker.worker_id, now=current)]
@@ -0,0 +1,84 @@
"""Privacy, artifact, and runtime health helpers for V2R1C7."""
from __future__ import annotations
import argparse
import hashlib
import json
import re
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from pathlib import Path
from typing import Any, Mapping
from deployment_state import RuntimeDeploymentState
_SECRET_KEY = re.compile(r"(?i)(?:token|secret|password|private[_-]?key|api[_-]?key)")
_SECRET_VALUE = re.compile(r"(?i)(?:bearer\s+)[A-Za-z0-9._~+/=-]+|(?:token|secret|password|private[_-]?key|api[_-]?key)\s*[:=]\s*[^\s,;]+")
def redact_text(value: str) -> str:
if not isinstance(value, str): raise TypeError("text_required")
return _SECRET_VALUE.sub("[REDACTED]", value)
def redact_value(value: Any, *, key: str = "") -> Any:
if isinstance(value, Mapping):
return {str(k): ("[REDACTED]" if _SECRET_KEY.search(str(k)) else redact_value(v, key=str(k))) for k,v in value.items()}
if isinstance(value, list): return [redact_value(v, key=key) for v in value]
if isinstance(value, tuple): return [redact_value(v, key=key) for v in value]
if isinstance(value, str): return "[REDACTED]" if _SECRET_KEY.search(key) else redact_text(value)
return value
def sha256_file(path: Path) -> str:
if path.is_symlink() or not path.is_file(): raise ValueError("regular_file_required")
digest=hashlib.sha256()
with path.open("rb") as stream:
for chunk in iter(lambda: stream.read(1024*1024), b""): digest.update(chunk)
return digest.hexdigest()
def _safe_relative(root: Path, name: str) -> Path:
relative=Path(name)
if relative.is_absolute() or ".." in relative.parts or name in {"", "."}: raise ValueError("unsafe_manifest_path")
candidate=(root/relative).resolve(); candidate.relative_to(root.resolve()); return candidate
def verify_manifest(root: Path, manifest: Path | None = None) -> dict[str, Any]:
root=root.resolve(); manifest=manifest or root/"artifacts"/"MANIFEST.sha256"
if manifest.is_symlink() or not manifest.is_file(): raise ValueError("manifest_missing")
checked=0
for raw in manifest.read_text(encoding="utf-8").splitlines():
line=raw.strip()
if not line: continue
parts=line.split(maxsplit=1)
if len(parts)!=2 or not re.fullmatch(r"[0-9a-f]{64}", parts[0]): raise ValueError("manifest_entry_invalid")
expected,name=parts; name=name[1:] if name.startswith("*") else name
if sha256_file(_safe_relative(root,name))!=expected: raise ValueError("manifest_hash_mismatch")
checked+=1
if not checked: raise ValueError("manifest_empty")
return {"status":"PASS","entries":checked}
def load_release_metadata(path: Path) -> dict[str, Any]:
metadata=json.loads(path.read_text(encoding="utf-8"))
if metadata.get("version") not in {"V2R1C6","V2R1C7"}: raise ValueError("version_mismatch")
if any(metadata.get(k) is not False for k in ("production_deployed","tag_created","baseline_changed")): raise ValueError("production_change_not_allowed")
return metadata
def health_payload(state_provider: RuntimeDeploymentState | None = None) -> dict[str, Any]:
return (state_provider or RuntimeDeploymentState.from_environment()).health_payload()
class _HealthHandler(BaseHTTPRequestHandler):
def do_GET(self) -> None:
if self.path not in {"/","/health","/healthz"}: self.send_error(404); return
body=json.dumps(health_payload(), sort_keys=True).encode()
self.send_response(200); self.send_header("Content-Type","application/json"); self.send_header("Content-Length",str(len(body))); self.end_headers(); self.wfile.write(body)
def log_message(self, _format: str, *_args: object) -> None: return
def serve(host: str, port: int) -> None:
if host not in {"127.0.0.1","::1","localhost"}: raise ValueError("loopback_only_listener_required")
if not 1024 <= port <= 65535: raise ValueError("invalid_listener_port")
server=ThreadingHTTPServer((host,port),_HealthHandler)
try: server.serve_forever()
finally: server.server_close()
def main() -> int:
parser=argparse.ArgumentParser(); parser.add_argument("--root",type=Path,default=Path(".")); parser.add_argument("--manifest",type=Path); parser.add_argument("--serve",action="store_true"); parser.add_argument("--host",default="127.0.0.1"); parser.add_argument("--port",type=int,default=18927); args=parser.parse_args()
if args.serve: serve(args.host,args.port); return 0
print(json.dumps(verify_manifest(args.root,args.manifest),sort_keys=True)); return 0
if __name__ == "__main__": raise SystemExit(main())
@@ -0,0 +1,5 @@
"""V2R1C7 production health entry point; state is install/runtime supplied."""
from v2r1c6_privacy_tools import main
if __name__ == "__main__":
raise SystemExit(main())
@@ -0,0 +1,23 @@
{
"schema": "aios.release-approval-supersession.v1",
"version": "V2R1C6",
"status": "SUPERSEDED_PREDEPLOY_VALIDATION_FAILED",
"execute_allowed": false,
"old_binding": {
"target_oid": "28081c5522a869762ec4946846f68047d53d99f0",
"evidence_closure_oid": "fed41014d80bacca70c8f309a5fb2c22e6001e32",
"artifact_sha256": "437b1f891f23a1dd12feabeced3edee3c54cc552d990d6d852805c8d5a165150",
"approval_plan_id": "sha256:596c8d40c25f9427c58cbdb0bbae58c03ba5c200fec32025c22d8004d27d8a51"
},
"reasons": [
"artifact_candidate_descriptor",
"deployable_package_not_verified",
"production_controller_support_not_verified"
],
"replacement_required": true,
"production_deployed": false,
"tag_created": false,
"baseline_changed": false,
"superseded_at": "2026-08-07",
"note": "Retained for audit; never delete, overwrite, or reuse."
}
@@ -0,0 +1,9 @@
{
"schema": "aios.approval.supersession.v1",
"version": "V2R1C7",
"superseded_approval_plan_id": "sha256:e62afc9946bcd0badc8e75291baf979e915dccaa50c2c77912c77f1ddeb2aa65",
"status": "SUPERSEDED_PREMERGE_IDENTITY_CHANGED",
"execute": false,
"reason": "release identity and artifact source head changed before protected merge",
"production_deployed": false
}
+14
View File
@@ -0,0 +1,14 @@
26d56250af61748d26a75fb35aaaf46578d0520fc11796b75de2330ef2ebb64f VERSION
81afc7ed5c1123146f61cf7bcf96c1fb69baa9371f1beec693341f20d612e0d4 active-runtime/app/v2r1c6_privacy_tools.py
437b1f891f23a1dd12feabeced3edee3c54cc552d990d6d852805c8d5a165150 artifacts/v2r1c6-candidate-artifact.txt
9f0dced0b252d2103ac4e975e17fe220aa990d4e3ebcfa17ad3ec32cf0ce4152 deploy/aios-v2/systemd/aios-v2r1c6-production.env.example
cee41f7c6e5a56632c6cb015d9182a6376d502fbf7a53b701369bf062c33f359 deploy/aios-v2/systemd/aios-v2r1c6-production.service
3f050310303662eb71840c0cb7cac5a5523968750ea1dd428653d5ba1ef2b727 deploy/aios-v2/v2r1c6-rollback.json
cbfefff50c6b1a6bc98de138e468929b4a98b738baeb0e0c97d2672f13ccc452 docs/releases/V2R1C6-CANDIDATE.md
a3813d060d842c22ff1a0f5ed968625b4d9d878fbacccb76d924f365d1c1a3a5 scripts/verify_v2r1c6_candidate.py
a194638a44668617ddd2c12e45a43b7f0020d5f0308fcadd9f8da1db5d6bb645 scripts/verify_v2r1c6_candidate.sh
951254504f20ae8d0d2397c2c2b2baeb2127651eb1e6cd1dc06695b836e176f9 tests/aios_v2_candidate/conftest.py
43322bdfb3cffa9d4756d929741f942a0da45b2498c7b5ed91906e115d968948 tests/aios_v2_candidate/test_v2r1c5_privacy_tools.py
068a35099d17047d127ac06be1a84adf090976fab13cf321febb28c787c0cac5 tests/aios_v2_candidate/test_v2r1c6_privacy_tools.py
82ab1aa2d50d74a96392a01bb3c0b80eb9033f5bcbd5555487c22bc9250f412d tests/aios_v2_candidate/test_v2r1c6_release_consistency.py
3b4a466cee373714805f50232d129ba06322027c51c467d1f35ab2c852b64c12 versions/V2R1C6.json
+8
View File
@@ -0,0 +1,8 @@
AIOS V2R1C6 candidate artifact
repository: aios/aios
base_ref: refs/heads/main
base_oid: a33e2593967a180c89d44454350ccaeaf3f79b25
candidate_branch: agent/v2r1c6-mainline-reconciliation-20260806
production_deployed: false
tag_created: false
baseline_changed: false
+1
View File
@@ -0,0 +1 @@
{"artifact":"aios-V2R1C7-92b4a937e56b.tar.zst","artifact_sha256":"6c7c763dc9a63a8411be3fc458881fc9bca1f403355d2f3a1cb338edbf9f24da","artifact_size":9672,"base_oid":"b356dec2ac5bd2a5f721aa73f8fd1902f9c65632","baseline_changed":false,"deployable":true,"executable_payload":true,"production_deployed":false,"schema":"aios.release-artifact-receipt.v3","source_head_oid":"7f09670511138f1f1f74f6b40329cd35d6eb2e55","tag_created":false,"version":"V2R1C7"}
+79
View File
@@ -0,0 +1,79 @@
{
"duplicates": 0,
"files": [
{
"path": "VERSION",
"sha256": "9208325e5b2e6f5ba6b64e000bc21d14d9f2f214e3ad33dd79857e85218fd6c2",
"size": 7
},
{
"path": "versions/V2R1C7.json",
"sha256": "8f3153c625241f60b65109eeac045e8aed7661d95f3a10f0b32ba3e0e9ac0a85",
"size": 801
},
{
"path": "release-manifests/V2R1C7.json",
"sha256": "560e5d9af8a8e2ec60ec31cf8659a7982b93ea09b7adb486b738ea2fba460d70",
"size": 735
},
{
"path": "deploy/aios-v2/systemd/aios-v2r1c7-production.service",
"sha256": "c96ba560d69f6c2eb700b9f2b42a32e394dadd00dffb23bbefdb9fb3e1571c0d",
"size": 709
},
{
"path": "active-runtime/app/deployment_state.py",
"sha256": "0997ab8332985c7124f334c4e49735d8833326578acc61a61a389bf792424de3",
"size": 2657
},
{
"path": "active-runtime/app/v2r1c6_privacy_tools.py",
"sha256": "dd3f8b1cab88e005a291f077807873c381e8ca92fe863dc05817925f0dff9613",
"size": 4603
},
{
"path": "active-runtime/app/v2r1c7_health_service.py",
"sha256": "9e45025c41068cafdea2f360edfc4146a5459b0a6bcfbf1403206fc0161e7bac",
"size": 174
},
{
"path": "scripts/build_v2r1c7_release_artifact.py",
"sha256": "81a423c7814f81cb888b60a538c153e549b2be9099c881c968d55ae84efddc57",
"size": 4787
},
{
"path": "scripts/verify_v2r1c7_release.py",
"sha256": "32a860e28d5afc105f413d878a13fb8600fa74874dbd3bc4c031ef3efa17896a",
"size": 1228
},
{
"path": "scripts/probe_v2r1c7_health.py",
"sha256": "64c78dabffb05c8851f87b2a27c43bd1f21728dca586397d03a6bc22906a906c",
"size": 681
},
{
"path": "scripts/rollback_v2r1c7_release.sh",
"sha256": "5419dc7de0f50908ad0ef1dac74172fce811bb7b45f314a6e1c1bb9ce618324d",
"size": 97
}
],
"mismatched": 0,
"missing": 0,
"missing_paths": [],
"required_files": [
"VERSION",
"versions/V2R1C7.json",
"release-manifests/V2R1C7.json",
"deploy/aios-v2/systemd/aios-v2r1c7-production.service",
"active-runtime/app/deployment_state.py",
"active-runtime/app/v2r1c6_privacy_tools.py",
"active-runtime/app/v2r1c7_health_service.py",
"scripts/build_v2r1c7_release_artifact.py",
"scripts/verify_v2r1c7_release.py",
"scripts/probe_v2r1c7_health.py",
"scripts/rollback_v2r1c7_release.sh"
],
"schema": "aios.v2r1c7.source-manifest.v1",
"unexpected": 0,
"version": "V2R1C7"
}
@@ -0,0 +1 @@
{"artifact":"aios-V2R1C7-92b4a937e56b.tar.zst","artifact_sha256":"6c7c763dc9a63a8411be3fc458881fc9bca1f403355d2f3a1cb338edbf9f24da","artifact_size":9672,"base_oid":"b356dec2ac5bd2a5f721aa73f8fd1902f9c65632","baseline_changed":false,"deployable":true,"executable_payload":true,"production_deployed":false,"schema":"aios.release-artifact-receipt.v3","source_head_oid":"7f09670511138f1f1f74f6b40329cd35d6eb2e55","tag_created":false,"version":"V2R1C7"}
Binary file not shown.
@@ -0,0 +1 @@
{"artifact":"aios-V2R1C7-v2r1c7.tar.zst","artifact_sha256":"08864f3f2d7e674a908a1d6eb7436db8747bd917200bf206b01276ec3da1e4be","artifact_size":7993,"baseline_changed":false,"deployable":true,"executable_payload":true,"production_deployed":false,"schema":"aios.release-artifact-receipt.v2","source_base_oid":"b356dec2ac5bd2a5f721aa73f8fd1902f9c65632","tag_created":false,"version":"V2R1C7"}
Binary file not shown.
+72 -5
View File
@@ -1,5 +1,5 @@
{
"schema_version": 1,
"schema_version": 4,
"automatic": [
"build",
"test",
@@ -8,8 +8,7 @@
"evidence"
],
"owner": [
"production_runtime_switch",
"production_tag"
"signed_release_plan"
],
"forbidden": [
"force_push",
@@ -19,7 +18,8 @@
"registry_mutation"
],
"runtime_deployment": {
"approval_action": "production_runtime_switch",
"approval_action": "signed_release_plan",
"required_capability": "runtime_switch",
"tag_operations_allowed": false,
"baseline_mutation_allowed": false,
"registry_mutation_allowed": false,
@@ -27,10 +27,77 @@
"rollback_required": true
},
"tag_release": {
"approval_action": "production_tag",
"approval_action": "signed_release_plan",
"required_capability": "tag_publish",
"separate_controller": true,
"duplicate_conflict": "fail_closed",
"rewrite_allowed": false,
"force_push_allowed": false
},
"baseline_registry_finalize": {
"approval_action": "signed_release_plan",
"required_capability": "baseline_registry_finalize",
"separate_controller": true,
"remote_tag_verification_required": true,
"gitea_api_verification_required": true,
"registry_baseline_consistency_required": true
},
"approval": {
"algorithm": "Ed25519",
"trust_domain": "aios.production",
"maximum_lifetime_seconds": 3600,
"release_plan_sha256_required": true,
"not_before_required": true,
"replay_protection": true,
"revocation_fail_closed": true,
"key_rotation_supported": true,
"capabilities": [
"runtime_switch",
"tag_publish",
"baseline_registry_finalize"
]
},
"execution": {
"environment_variable": "AIOS_SILENT_RELEASE_EXECUTION_ENABLED",
"default_enabled": false,
"shadow_side_effects_allowed": false,
"durable_saga_required": true,
"pre_switch_revalidation_required": true,
"trusted_candidate_builder_required": true,
"actual_plan_regeneration_required": true,
"startup_fail_closed_on_missing_component": true,
"recovery_worker_shadow_default": true,
"production_approval_backend_required": true,
"production_builder_signer": "kms-hsm-secret-manager-only",
"production_execution_without_backend": "unavailable"
},
"durability": {
"journal": "sha256-hash-chain",
"probe_journal": "sha256-hash-chain",
"exclusive_capability_claims": true,
"side_effect_probe_before_lease_takeover": true,
"side_effect_probe_before_failed_recoverable_retry": true,
"probe_results": [
"NOT_APPLIED",
"APPLIED_VERIFIED",
"INDETERMINATE",
"CONFLICT"
],
"probe_operation_digest_required": true,
"probe_freshness_required": true,
"tamper_result": "MANUAL_OWNER_REVIEW_REQUIRED"
},
"tag_pending": {
"required_observations": [
"remote_git_ref",
"annotated_tag_object",
"peeled_target",
"gitea_api",
"gitea_listing",
"repository_identity"
],
"indeterminate_result": "RECONCILIATION_REQUIRED",
"conflict_result": "MANUAL_OWNER_REVIEW_REQUIRED",
"automatic_reconciliation": false
}
}
@@ -0,0 +1,4 @@
AIOS_V2_VERSION=V2R1C6
AIOS_V2R1C6_PORT=18926
AIOS_GITOPS_WRITES_ENABLED=false
AIOS_V2R1C6_PRODUCTION_DEPLOYED=false
@@ -0,0 +1,30 @@
[Unit]
Description=AIOS V2R1C6 candidate privacy service (not production-enabled)
After=network.target
[Service]
Type=simple
User=aios-v2r1c6
Group=aios-v2r1c6
WorkingDirectory=/opt/aios/v2r1c6
EnvironmentFile=-/etc/aios/aios-v2r1c6.env
Environment=AIOS_GITOPS_WRITES_ENABLED=false
Environment=AIOS_V2R1C6_PRODUCTION_DEPLOYED=false
ExecStart=/usr/bin/python3 /opt/aios/v2r1c6/active-runtime/app/v2r1c6_privacy_tools.py --serve --host 127.0.0.1 --port 18926
Restart=on-failure
RestartSec=5
NoNewPrivileges=true
PrivateTmp=true
ProtectHome=true
ProtectSystem=strict
ProtectKernelLogs=true
ProtectKernelModules=true
ProtectKernelTunables=true
ProtectControlGroups=true
RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6
IPAddressDeny=any
IPAddressAllow=localhost
UMask=0077
[Install]
WantedBy=multi-user.target
@@ -0,0 +1,8 @@
# Installer generates these bindings; STAGED is the only pre-acceptance state.
RELEASE_VERSION=V2R1C7
TARGET_OID=__INSTALL_TARGET_OID__
ARTIFACT_SHA256=__INSTALL_ARTIFACT_SHA256__
DEPLOYMENT_STATE=STAGED
AIOS_V2R1C7_PORT=18927
DEPLOYMENT_STATE_FILE=/var/lib/aios/v2r1c7/deployment-state.json
AIOS_GITOPS_WRITES_ENABLED=false
@@ -0,0 +1,28 @@
[Unit]
Description=AIOS V2R1C7 governed health contract service
After=network.target
[Service]
Type=simple
User=aios-v2r1c7
Group=aios-v2r1c7
WorkingDirectory=/opt/aios/v2r1c7
EnvironmentFile=/etc/aios/aios-v2r1c7.env
ExecStart=/usr/bin/python3 /opt/aios/v2r1c7/active-runtime/app/v2r1c7_health_service.py --serve --host 127.0.0.1 --port ${AIOS_V2R1C7_PORT}
Restart=on-failure
RestartSec=5
NoNewPrivileges=true
PrivateTmp=true
ProtectHome=true
ProtectSystem=strict
ProtectKernelLogs=true
ProtectKernelModules=true
ProtectKernelTunables=true
ProtectControlGroups=true
RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6
IPAddressDeny=any
IPAddressAllow=localhost
UMask=0077
[Install]
WantedBy=multi-user.target
+17
View File
@@ -0,0 +1,17 @@
{
"schema": "aios.v2r1c6.candidate-rollback.v1",
"candidate_version": "V2R1C6",
"rollback_version": "V2R1C5",
"previous_tag": "V2R1C5",
"previous_tag_oid": "9e65bd59baea392294c0f253046d3ab116f63145",
"previous_target_oid": "aa60e891547abd7e215af7cf2bf3f21fd23541da",
"baseline_unchanged": true,
"tag_rewrite": false,
"force_push": false,
"production_deploy_authorized": false,
"rollback_steps": [
"stop aios-v2r1c6-production.service if a separately approved deployment exists",
"restore the V2R1C5 production artifact and service configuration",
"verify V2R1C5 tag and Baseline before any restart"
]
}
+9
View File
@@ -0,0 +1,9 @@
{
"schema": "aios.release-rollback.v1",
"version": "V2R1C7",
"previous_version": "V2R1C6",
"previous_tag": "V2R1C6",
"rollback_state": "ROLLED_BACK",
"production_deploy_authorized": false,
"execution": "manual_owner_approval_required"
}
@@ -0,0 +1,8 @@
{
"schema": "aios.v2r1c7.runtime-config.v1",
"install_state": "STAGED",
"acceptance_transition": "STAGED -> ACTIVE only after staging gates",
"production_deploy_authorized": false,
"registry_write_authorized": false,
"baseline_write_authorized": false
}
@@ -0,0 +1,24 @@
[Unit]
Description=AIOS manifest-driven release controller (validation only)
ConditionPathExists=/var/lib/aios-release/release-source-manifest.json
ConditionPathExists=/var/lib/aios-release/release-request.json
After=network-online.target
[Service]
Type=oneshot
User=aios-release
Group=aios-release
Environment=PYTHONPATH=/opt/aios/release-controller
Environment=AIOS_GITOPS_WRITES_ENABLED=false
Environment=AIOS_PRODUCTION_DEPLOYMENT_AUTHORIZED=false
ExecStart=/usr/bin/python3 -m aios_release_controller.cli validate-release /var/lib/aios-release/release-source-manifest.json /var/lib/aios-release/release-request.json
NoNewPrivileges=yes
PrivateTmp=yes
ProtectSystem=strict
ProtectHome=yes
ReadWritePaths=/var/lib/aios-release
CapabilityBoundingSet=
RestrictAddressFamilies=AF_UNIX
[Install]
WantedBy=multi-user.target
@@ -0,0 +1,24 @@
[Unit]
Description=AIOS governed Agent Pull Runner (example)
After=network.target
[Service]
Type=simple
User=aios-agent
Group=aios-agent
UMask=0077
ExecStart=/usr/bin/python3 -B /opt/aios-gitops/active-runtime/app/agent_pull_runner_cli.py --worker-id %H
NoNewPrivileges=true
PrivateTmp=true
PrivateDevices=true
ProtectSystem=strict
ProtectHome=true
ProtectKernelTunables=true
ProtectKernelModules=true
ProtectControlGroups=true
RestrictSUIDSGID=true
RestrictAddressFamilies=AF_UNIX AF_INET
ReadWritePaths=/var/lib/aios-gitops/runtime/agent-pull-runner
Restart=on-failure
RestartSec=5
@@ -0,0 +1,42 @@
[Unit]
Description=AIOS Autonomous Release Candidate API (Dormant/Shadow Only)
After=network.target
ConditionPathExists=/opt/aios-autonomous-release-candidate/aios_autonomous_release_candidate
[Service]
Type=simple
User=aiman
Group=gitops
UMask=0077
WorkingDirectory=/opt/aios-autonomous-release-candidate
Environment=PYTHONDONTWRITEBYTECODE=1
Environment=PYTHONPATH=/opt/aios-autonomous-release-candidate
Environment=AIOS_TASK_PLANE_EXECUTION_ENABLED=false
Environment=AIOS_GITOPS_WRITES_ENABLED=false
Environment=AIOS_PRODUCTION_DEPLOYMENT_ENABLED=false
ExecStart=/usr/bin/python3 -m aios_autonomous_release_candidate.service --host 127.0.0.1 --port 19951 --state /var/lib/aios-autonomous-release-candidate/state.sqlite3
StateDirectory=aios-autonomous-release-candidate
RuntimeDirectory=aios-autonomous-release-candidate
NoNewPrivileges=true
PrivateTmp=true
PrivateDevices=true
ProtectSystem=strict
ProtectHome=true
ProtectKernelTunables=true
ProtectKernelModules=true
ProtectKernelLogs=true
ProtectControlGroups=true
ProtectClock=true
ProtectHostname=true
LockPersonality=true
RestrictSUIDSGID=true
RestrictRealtime=true
MemoryDenyWriteExecute=true
RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6
IPAddressDeny=any
IPAddressAllow=localhost
ReadOnlyPaths=/opt/aios-autonomous-release-candidate
ReadWritePaths=/var/lib/aios-autonomous-release-candidate
[Install]
WantedBy=multi-user.target
@@ -0,0 +1,51 @@
[Unit]
Description=AIOS Autonomous Release Candidate Worker (Dormant/Shadow Only)
After=network.target
ConditionPathExists=/opt/aios-autonomous-release-candidate/aios_autonomous_release_candidate
[Service]
Type=simple
User=aiman
Group=gitops
UMask=0077
WorkingDirectory=/opt/aios-autonomous-release-candidate
Environment=PYTHONDONTWRITEBYTECODE=1
Environment=PYTHONPATH=/opt/aios-autonomous-release-candidate
Environment=AIOS_TASK_PLANE_EXECUTION_ENABLED=false
Environment=AIOS_GITOPS_WRITES_ENABLED=false
Environment=AIOS_PRODUCTION_DEPLOYMENT_ENABLED=false
Environment=AIOS_CANDIDATE_CONTROL_ENDPOINT=127.0.0.1:19951
Environment=AIOS_OPENCLAW_CREDENTIAL_REFERENCE=systemd-credential:openclaw-gateway-candidate
Environment=AIOS_CODEX_CREDENTIAL_REFERENCE=systemd-credential:codex-app-server-candidate
Environment=AIOS_NATIVE_CONTROL_CREDENTIAL_REFERENCE=systemd-credential:session-native-adapter-control-candidate
Environment=AIOS_GITEA_CREDENTIAL_REFERENCE=systemd-credential:gitea-admin-api-candidate
ImportCredential=openclaw-gateway-candidate
ImportCredential=codex-app-server-candidate
ImportCredential=session-native-adapter-control-candidate
ImportCredential=gitea-admin-api-candidate
ExecStart=/usr/bin/python3 -m aios_autonomous_release_candidate.worker_service --state-directory /var/lib/aios-autonomous-release-worker-candidate
StateDirectory=aios-autonomous-release-worker-candidate
RuntimeDirectory=aios-autonomous-release-worker-candidate
NoNewPrivileges=true
PrivateTmp=true
PrivateDevices=true
ProtectSystem=strict
ProtectHome=true
ProtectKernelTunables=true
ProtectKernelModules=true
ProtectKernelLogs=true
ProtectControlGroups=true
ProtectClock=true
ProtectHostname=true
LockPersonality=true
RestrictSUIDSGID=true
RestrictRealtime=true
MemoryDenyWriteExecute=true
RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6
IPAddressDeny=any
IPAddressAllow=localhost
ReadOnlyPaths=/opt/aios-autonomous-release-candidate
ReadWritePaths=/var/lib/aios-autonomous-release-worker-candidate
[Install]
WantedBy=multi-user.target
@@ -0,0 +1,16 @@
[Unit]
Description=AIOS branch-bound repair lease control plane
After=network.target
[Service]
Type=exec
User=root
Group=root
UMask=0077
Environment=PYTHONDONTWRITEBYTECODE=1
ExecStart=/usr/bin/python3 -B /opt/aios-gitops/active-runtime/app/aios_branch_repair_lease_api.py
Restart=on-failure
NoNewPrivileges=no
# Bind the transport through the existing authenticated loopback control plane.
# This service never owns a Git ref and never receives bearer plaintext.
@@ -0,0 +1,15 @@
# Template only. Deployment Agent must adapt the existing absolute application path.
# /etc/systemd/system/aios-gitops-model-api.service.d/40-authorized-writer-bridge.conf
[Service]
Environment=AIOS_GITOPS_REPOSITORY_IDENTITY=aios/aios
Environment=AIOS_GITOPS_AUTHORITY_GIT_DIR=/data/aios-gitops/repo.git
Environment=AIOS_GITOPS_AUTHORIZED_WRITER_STATE_ROOT=/var/lib/aios-gitops/authorized-writer-bridge
Environment=AIOS_GITOPS_GRANT_SIGNER_ADAPTER=/usr/local/libexec/aios-grant-signer-adapter
Environment=AIOS_GITOPS_GRANT_VALIDATOR=/usr/local/libexec/aios-grant-validator
Environment=AIOS_GITOPS_TASK_WRITER_PROVISIONER=/usr/local/libexec/aios-task-writer-provisioner
Environment=AIOS_GITOPS_OWNER_PROVISION_SCOPE=writer:provision
Environment=AIOS_GITOPS_GOVERNED_USE_SUDO=true
Environment=AIOS_GITOPS_SUDO_PATH=/usr/bin/sudo
# Replace the existing Python entrypoint only after staging acceptance.
# ExecStart=
# ExecStart=/usr/bin/python3 /opt/aios/active-runtime/app/aios_authorized_task_writer_api.py
@@ -0,0 +1,34 @@
# Template only. Deployment Agent must substitute non-secret, verified values.
# /etc/systemd/system/aios-gitops-model-api.service.d/45-owner-authorization-broker.conf
[Service]
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_BROKER_ENABLED=true
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_STATE_ROOT=/var/lib/aios-gitops/owner-authorization-broker
# Select exactly one scope already present on the intended non-wildcard Owner
# credential. This is only an authentication intake scope; it is not copied
# into the generated task-writer grant.
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_INTAKE_SCOPES=commit:create
# Required explicit Owner account allowlist. Replace with the verified account
# from the existing token registry; never put the bearer token here.
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_ALLOWED_IDENTITIES=REPLACE_WITH_OWNER_ACCOUNT
# If handler.need() does not return non-secret scope metadata, configure the
# exact non-secret credential ID instead. Leave empty when scope readback is
# available. Never put a token or token digest in this value.
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_ALLOWED_CREDENTIAL_IDS=
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_MAX_EXPIRY_SECONDS=900
Environment=AIOS_GITOPS_REPOSITORY_IDENTITY=aios/aios
Environment=AIOS_GITOPS_AUTHORITY_GIT_DIR=/data/aios-gitops/repo.git
Environment=AIOS_GITOPS_AUTHORIZED_WRITER_STATE_ROOT=/var/lib/aios-gitops/authorized-writer-bridge
Environment=AIOS_GITOPS_GRANT_SIGNER_ADAPTER=/usr/local/libexec/aios-grant-signer-adapter
Environment=AIOS_GITOPS_GRANT_VALIDATOR=/usr/local/libexec/aios-grant-validator
Environment=AIOS_GITOPS_TASK_WRITER_PROVISIONER=/usr/local/libexec/aios-task-writer-provisioner
Environment=AIOS_GITOPS_GOVERNED_USE_SUDO=true
Environment=AIOS_GITOPS_SUDO_PATH=/usr/bin/sudo
# Replace the current Python entrypoint only after staging acceptance.
# This wrapper preserves existing routes and both bridge endpoints.
# ExecStart=
# ExecStart=/usr/bin/python3 /opt/aios/active-runtime/app/aios_owner_authorization_broker_api.py
@@ -0,0 +1,16 @@
[Unit]
Description=AIOS temporary governed local writer
Requires=aios-governed-local-writer.socket
After=aios-governed-local-writer.socket
[Service]
Type=simple
User=root
Group=root
UMask=0077
Environment=AIOS_GITOPS_WRITES_ENABLED=true
Environment=PYTHONDONTWRITEBYTECODE=1
ExecStart=/usr/bin/python3 -B /opt/aios-gitops/active-runtime/app/aios_governed_local_writer_service.py
Restart=no
PrivateNetwork=yes
NoNewPrivileges=no
@@ -0,0 +1,13 @@
[Unit]
Description=AIOS governed local writer Unix socket
[Socket]
ListenStream=/run/aios-gitops/governed-local-writer.sock
SocketUser=root
SocketGroup=root
SocketMode=0600
DirectoryMode=0700
RemoveOnStop=yes
[Install]
WantedBy=sockets.target
@@ -0,0 +1,25 @@
[Unit]
Description=AIOS Silent Release Shadow Controller
After=network-online.target
Wants=network-online.target
[Service]
Type=simple
User=aiman
Group=gitops
UMask=0077
Environment=AIOS_GITOPS_WRITES_ENABLED=false
Environment=AIOS_SILENT_RELEASE_EXECUTION_ENABLED=false
Environment=PYTHONDONTWRITEBYTECODE=1
ExecStart=/usr/bin/python3 /usr/libexec/aios-silent-release-shadow/current/active-runtime/app/silent_release_execution.py --state-root /var/lib/aios-silent-release-shadow --repository /data/aios-gitops/repo.git --host 127.0.0.1 --port 18891 --poll-seconds 30
Restart=on-failure
RestartSec=5
NoNewPrivileges=true
PrivateTmp=true
ProtectSystem=strict
ProtectHome=true
ReadWritePaths=/var/lib/aios-silent-release-shadow
ReadOnlyPaths=/data/aios-gitops/repo.git /usr/libexec/aios-silent-release-shadow
[Install]
WantedBy=multi-user.target
@@ -0,0 +1,59 @@
# Staging-only task identity overlay for V2R1C6 bootstrap.
# Install as /etc/systemd/system/aios-v2r1c6-task-identity-overlay.service
# after substituting the verified staging path.
[Unit]
Description=AIOS V2R1C6 task-scoped identity overlay
After=network.target
ConditionPathIsDirectory=/data/aios-gitops/repo.git
[Service]
Type=simple
User=root
Group=root
WorkingDirectory=REPLACE_WITH_ROOT_OWNED_STAGING_DIR
Environment=PYTHONPATH=REPLACE_WITH_ROOT_OWNED_STAGING_DIR/active-runtime/app
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_ENABLED=true
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_HOST=127.0.0.1
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_PORT=19864
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_STATE_ROOT=/var/lib/aios-gitops/task-identity-overlays
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_SECRET_ROOT=/run/aios-gitops/task-identity-overlays
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_TASK_ID=V2R1C6_TASK_SCOPED_WRITER_PROVISIONING
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_OWNER_IDENTITY=aiman
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_BASE_REF=refs/heads/main
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_BASE_OID=875d41c3101093149414292067fd700dc214d4d3
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_TARGET_BRANCH=agent/v2r1c6-mainline-reconciliation-20260805
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_MAX_TTL_SECONDS=900
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_REQUIRE_ROOT=true
# Agent must enumerate every active bearer secret reference discovered on-host.
Environment=AIOS_GITOPS_IDENTITY_OVERLAY_COLLISION_SECRET_REFS=REPLACE_WITH_COMMA_SEPARATED_ACTIVE_SECRET_REFS
Environment=AIOS_GITOPS_REPOSITORY_IDENTITY=aios/aios
Environment=AIOS_GITOPS_AUTHORITY_GIT_DIR=/data/aios-gitops/repo.git
Environment=AIOS_GITOPS_OWNER_AUTHORIZATION_STATE_ROOT=/var/lib/aios-gitops/owner-authorization-broker
Environment=AIOS_GITOPS_AUTHORIZED_WRITER_STATE_ROOT=/var/lib/aios-gitops/authorized-writer-bridge
Environment=AIOS_GITOPS_GRANT_SIGNER_ADAPTER=/usr/local/libexec/aios-grant-signer-adapter
Environment=AIOS_GITOPS_GRANT_VALIDATOR=/usr/local/libexec/aios-grant-validator
Environment=AIOS_GITOPS_TASK_WRITER_PROVISIONER=/usr/local/libexec/aios-task-writer-provisioner
Environment=AIOS_GITOPS_GOVERNED_USE_SUDO=false
ExecStart=/usr/bin/python3 REPLACE_WITH_ROOT_OWNED_STAGING_DIR/active-runtime/app/aios_task_identity_overlay_api.py
Restart=no
RuntimeDirectory=aios-gitops/task-identity-overlays
RuntimeDirectoryMode=0700
StateDirectory=aios-gitops/task-identity-overlays
StateDirectoryMode=0700
UMask=0077
NoNewPrivileges=true
PrivateTmp=true
ProtectHome=true
ProtectHostname=true
ProtectKernelLogs=true
ProtectKernelModules=true
ProtectKernelTunables=true
ProtectControlGroups=true
RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6
RestrictNamespaces=true
LockPersonality=true
MemoryDenyWriteExecute=true
SystemCallArchitectures=native
[Install]
WantedBy=multi-user.target
@@ -0,0 +1,33 @@
{
"action": "AUTHORIZE_TASK_WRITER_PROVISIONING",
"allowed_operations": [
"create_worktree",
"create_target_branch_if_absent",
"commit_to_target_branch",
"push_target_branch"
],
"base_oid": "875d41c3101093149414292067fd700dc214d4d3",
"base_ref": "refs/heads/main",
"baseline_change_authorized": false,
"denied_operations": [
"direct_main_push",
"force_push",
"tag_create",
"tag_update",
"tag_delete",
"branch_delete",
"repository_admin",
"baseline_write",
"production_deploy",
"production_marker_write",
"history_rewrite"
],
"expires_in_seconds": 900,
"main_write_authorized": false,
"one_time_use": true,
"production_change_authorized": false,
"repository": "aios/aios",
"tag_change_authorized": false,
"target_branch": "agent/v2r1c6-mainline-reconciliation-20260805",
"task_id": "V2R1C6_TASK_SCOPED_WRITER_PROVISIONING"
}
@@ -0,0 +1,71 @@
# AIOS Autonomous Release Agent Deployment Instructions
## Objective
Deploy only candidate/shadow environments first. Production activation requires a separate owner release approval.
## Agent execution order
1. Verify git references:
- local
- authoritative bare
- remote
- Gitea API
2. Build candidate artifact.
3. Run:
- unit tests
- security scans
- secret scans
- artifact verification
4. Start candidate workers:
- OpenClaw native adapter
- Codex native adapter
- dedicated writer
5. Execute shadow task:
```
TaskPlan
-> Planner
-> Worker Router
-> Session lifecycle
-> Evidence chain
-> Artifact verification
-> ReleasePlan
```
## Hard restrictions
Never:
- write main directly
- force push
- rewrite tags
- export secrets
- bypass owner approval
- modify production flags during candidate validation
## Production release gate
Require:
```
owner_approval=true
artifact_verified=true
evidence_closure=true
rollback_ready=true
remote_tag_verified=true
baseline_verified=true
registry_verified=true
```
Until then keep:
```
AIOS_GITOPS_WRITES_ENABLED=false
AIOS_TASK_PLANE_EXECUTION_ENABLED=false
AIOS_SILENT_RELEASE_EXECUTION_ENABLED=false
```
+34
View File
@@ -0,0 +1,34 @@
# AIOS Autonomous Release Candidate 架构
本 Candidate 把规划、工作流、凭据引用、原生会话、证据和发布控制拆成独立边界。它只负责验证约束,不拥有生产授权。
## 信任边界
- Owner Approval 的签名和撤销状态来自外部 HSM/KMS 或可信 Approval Plane。Candidate 只有 verifier,没有 signer。
- Secret Manager V2 只返回 `systemd-credential:*` 或受控的 Candidate reference,不返回认证正文。
- OpenClaw、Codex 和 Gitea provider 身份由管理面供应;模拟 provider 永远不能把原生 canary 标成已验证。
- Writer 只写任务绑定分支和允许路径。生产 main、tag、baseline、registry 和部署不属于 Candidate 权限。
## 核心模块
- `models.py`:TaskPlan、Approval、Credential、Result 和 ReleasePlan 强类型约束。
- `store.py`:私有 SQLite、WAL、全同步、schema 检查、事务和原子 step transition。
- `workflow.py`:依赖、claim、lease、retry、timeout recovery 和终态。
- `credential.py` / `secret_manager.py`:reference-only 生命周期、补偿撤销、跨任务拒绝和 post-revocation 验证。
- `native_adapters.py`:OpenClaw/Codex 原生 provider ID、parent/child tree、status、interrupt、close 和纯文本回执。
- `gitea_canary.py`:仅通过 deploy-key provider 的 list/create/delete 操作完成 canary 与撤销验证。
- `evidence.py`:落库的哈希链、重启回放、篡改检测和 required-event closure。
- `artifact.py`:确定性 tar、逐文件哈希、package closure、链接/特殊文件/敏感标记拒绝。
- `release.py` / `release_gate.py`:部署审批与回滚审批分离,tag object 与 peeled target 分离验证。
## 会话真实性
以下状态必须分别记录,不能互相替代:
```text
broker_session_verified
openclaw_native_child_session_verified
codex_child_session_verified
```
Shadow provider 只能证明 Broker 合同和清理路径,不能证明 OpenClaw 或 Codex 原生会话。
@@ -0,0 +1,47 @@
# AIOS Autonomous Release Execution Package
## Mode
Candidate/shadow first. Production changes require an independent release approval.
## Flow
1. Create TaskPlan.
2. Validate policy and approval envelope.
3. Route workers by capability and trust zone.
4. Execute on isolated candidate branch.
5. Collect evidence chain and artifact manifest.
6. Verify release plan.
7. Wait for owner approval.
8. Deploy only after approval.
## Hard stops
- main writes
- force push
- tag rewrite
- baseline mutation
- registry mutation
- plaintext credential handling
- missing evidence closure
## Required evidence
- git ref closure
- test report
- artifact checksum
- manifest checksum
- credential revoke proof
- production invariant snapshot
## Production activation
The executor must keep:
```
AIOS_GITOPS_WRITES_ENABLED=false
AIOS_TASK_PLANE_EXECUTION_ENABLED=false
AIOS_SILENT_RELEASE_EXECUTION_ENABLED=false
```
until release approval is independently verified.
+48
View File
@@ -0,0 +1,48 @@
# Candidate 运维与验收
## 安全默认值
两个 systemd 模板均为 Dormant/Shadow-only,绑定 loopback,默认关闭 Task Plane 执行、GitOps 写入和生产部署。模板进入仓库不等于安装或启动服务。
外部信任根缺失时必须保留以下阻断:
```text
BLOCKED_OWNER_APPROVAL_VERIFIER_UNAVAILABLE
BLOCKED_GITEA_ADMIN_CREDENTIAL_REQUIRED
BLOCKED_EXTERNAL_OWNER_SIGNATURE
```
不得用模拟身份、个人账户或生产 provider 身份替代这些根。
## 独立验收
从仓库根目录运行:
```text
PYTHONPATH=services/aios-autonomous-release-candidate \
python3 -m pytest -p no:cacheprovider -q tests/autonomous_release
PYTHONPATH=services/aios-autonomous-release-candidate \
python3 scripts/autonomous-release/run_acceptance.py \
--repo-root . \
--output /task-evidence/candidate-acceptance.json
```
Acceptance report 必须满足:
```text
all_passed=true
skipped=0
direct_gitea_database_access=false
production_path_hits=[]
secret_hits=[]
production_side_effects=false
```
## Artifact
Artifact builder 拒绝覆盖现有文件、链接、硬链接、特殊文件、仓库元数据和敏感内容标记。输出 tar 和 manifest 后必须再次执行 `ArtifactBuilder.verify`。
## 外部 canary
真实 native session 与 Gitea canary 只能在任务绑定的 runtime credential 投影存在时运行。每次 canary 完成后必须关闭 provider session、撤销上游对象、删除 projection,并保留非敏感回执。
+26
View File
@@ -0,0 +1,26 @@
# Governed ReleasePlan 关闭条件
ReleasePlan 必须绑定 task、repository、integration branch、完整 target/tree OID、artifact/manifest/evidence/rollback digest、生产配置 digest,以及互不相同的部署和回滚 ApprovalPlan ID。
## Candidate 与 Integration
1. Candidate 分支先完成测试、扫描、Shadow、Artifact 和外部门禁盘点。
2. 重新读取 authoritative `main` 完整 OID。
3. 从该 OID 创建 integration 分支。
4. 使用普通 `merge --no-ff` 合入 Candidate;禁止 squash、rebase、amend 和 force push。
5. 在 integration 上重新执行全部测试、扫描和 artifact closure。
## 生产停止线
下列项目任一为 false,流程必须停止在 `READY_FOR_OWNER_RELEASE_APPROVAL`,不得部署或创建生产 tag:
```text
openclaw_native_child_session_verified
codex_child_session_verified
gitea_admin_api_lifecycle_canary
owner_signer_separation_verified
credential_incident_closed
external_owner_release_approval_verified
```
聊天文本、Candidate 自签记录或模拟 canary 不能改变这条停止线。
+29
View File
@@ -0,0 +1,29 @@
# AIOS autonomy v1 and V2R1C6 governance components
This branch adds bounded control-plane primitives used by the V2R1C6 release
workflow. The components are deliberately transport-neutral; deployment
binds them to the existing task plane and governed local writer.
- `agent_pull_runner.py`: worker registration, heartbeat, long-poll with a
60-second fallback, jittered bounded backoff, claim/renew/report and restart
recovery with duplicate suppression.
- `autonomy_notification_plane.py`: SQLite durable outbox for Owner and GPT,
delivery retry and acknowledgement. Notification failure never rolls back
task execution.
- `recurring_task_scheduler.py`: interval scheduling with `next_run_at`, run
bounds, cancellation and idempotent successor creation.
- `release_state_machine.py`: ordered release states with fail-closed
production acceptance evidence.
- `reconciliation_controller.py`: read-only drift detection for refs, tags,
targets, registry/Baseline, artifact digest and service health.
- `governed_merge_gate.py`: supplements native Gitea protection when the
installed API cannot express Owner approval identity or conversation
resolution. It never writes refs and never authorizes force-push.
No component grants main, tag, Baseline, force-push, arbitrary-shell or
production-deploy capability.
The task writer persists request identity, idempotency keys, final commit OIDs,
parent OIDs and sanitized audit digests. A caller queries the receipt before
recovering an ambiguous request, and every branch update remains a
fast-forward CAS append.
+176 -207
View File
@@ -1,91 +1,80 @@
# AIOS Silent Release Controller Design Report
状态:isolated branch implementation
Production mutation:forbidden
目标流程:自动准备 → 自动验证 → 单次 Owner runtime 授权 → 自动部署/失败回滚
状态:隔离分支实现;生产 mutation 禁止
目标:自动准备 → 自动验证 → 单次 Owner 签名授权 → 分控制器自动执行与验收
## 1. 当前痛点
当前发布依赖多轮人工传递 commit、tree、archive、manifest、drop-in、测试和
canary 证据。证据在不同消息和报告中重复抄写,容易出现:
现有流程需要多轮人工传递 commit、tree、artifact、manifest、测试、canary、drop-in
和回滚证据。证据在不同消息里重复抄写,容易出现:
- artifact 在测试后被 cache 文件污染,closure 数量漂移;
- canary 通过,但批准文本与 runtime 实际字段契约不同;
- Git、Gitea、artifact、systemd 和 rollback 门禁由人手工拼接;
- runtime deployment 与 production tag 容易被误认为同一授权;
- 授权重放、失败回滚和持久化状态缺少统一接口。
- 构建后 artifact、manifest 或 policy 已变化,但旧批准仍被使用;
- Git、Gitea、systemd 与回滚结果由人工拼接,缺少同一一致性快照;
- runtime switch、tag、Baseline、Registry 的授权边界不清;
- 重启后只剩工作树内临时 JSON,无法可靠恢复;
- approval replay、并发 writer、key revoke 和 secret inheritance 缺少统一防线。
## 2. 新架构
```text
ReleaseBuilder adapter
Build / Test / Canary
|
v
SilentReleaseController.prepare(source_commit)
|
+--> ReleaseCandidateStore
ReleaseCandidate + DeploymentPlan
|
v
ReleaseAcceptanceEngine.evaluate(candidate)
|
+--> ReleaseProbe adapter
- local/bare/Gitea main
- repository identity
- artifact/manifest
- tests/canary
- rollback
- authoritative Git binding
ReleaseAcceptanceEngine
|
v
READY_FOR_APPROVAL
|
v one ApprovalEnvelope: production_runtime_switch
DeploymentController.deploy(candidate, approval)
v
SignedReleasePlanApproval (Ed25519)
|
+--> atomic switch
+--> restart only GitOps runtime
+--> health verification
+--> rollback on any failure
+--> runtime_switch capability
| DeploymentController
|
+--> tag_publish capability
| TagController
|
+--> baseline_registry_finalize capability
BaselineRegistryFinalizeController
TagController (separate seam)
|
+--> separate production_tag approval
+--> execute
+--> verify
Durable ReleaseSagaStore + CapabilityLedger + Event Journal
贯穿三个控制器;控制器互相不拥有对方的 mutation API。
```
`SilentReleaseController` 是外部深模块,只暴露 `prepare()` 和 `deploy()`。
构建、探针、systemd 与 tag provider 都是 adapter;复杂实现保持在模块内部。
`CommandReleaseBuilder` 使用无 shell 的 argv 调用确定性构建流水线,并只接受
与请求 source commit 一致的候选 JSON。
一次 Owner 批准是同一个不可变 release plan 的签名,不是一个万能执行入口。三个 capability
分别消费、分别留痕、分别幂等恢复。`force_push`、`tag_rewrite`、`direct_main_push`
永远不是可签发 capability。
## 3. 状态机
```text
BUILDING
|
v
VERIFYING ------------------------> FAILED
|
v
READY_FOR_APPROVAL --------------> FAILED
|
v
APPROVED ------------------------> FAILED
|
v
DEPLOYING -------> VERIFIED
| |
+-------------> ROLLED_BACK
|
+-------------> FAILED (rollback itself failed)
PREPARING
-> VERIFYING
-> READY_FOR_APPROVAL
-> APPROVED
-> PRE_SWITCH_REVALIDATING
-> SWITCH_PENDING
-> SWITCHED
-> RUNTIME_VERIFYING
-> TAG_PENDING
-> TAG_VERIFIED
-> BASELINE_REGISTRY_PENDING
-> COMPLETED
任一可恢复故障 -> RECONCILIATION_REQUIRED
对象冲突 -> MANUAL_OWNER_REVIEW_REQUIRED
切换后失败 -> ROLLBACK_PENDING -> ROLLED_BACK
```
状态不可跳跃。`VERIFIED` 只允许显式进入 `ROLLED_BACK`,终态不能自动回到准备态。
状态快照使用 revision compare-and-set;事件同时写入 append-only journal。writer 崩溃后根据
状态确定唯一恢复动作。并发控制器中只有一个 CAS winner。
## 4. 数据模型
`ReleaseCandidate`:
### ReleaseCandidate
```text
candidate_id
@@ -99,201 +88,181 @@ rollback_ready
git_consistency
gitea_consistency
approval_state
state
failure_reason
created_at
updated_at
```
核心 OID 和 SHA 在构造时验证。候选 fingerprint 绑定 candidate、commit、tree、
artifact 和 manifest,ApprovalEnvelope 必须逐字段匹配。
`AcceptanceEvidence` 是 adapter 与 acceptance engine 之间的 seam,包含三方 main、
repository identity、artifact/manifest、测试、canary、rollback 和 authoritative Git。
## 5. Interface
Python interface:
```python
candidate = controller.prepare(source_commit)
approval = ApprovalEnvelope.for_candidate(
candidate,
approval_id=...,
owner_id=...,
expires_at=...,
)
result = controller.deploy(candidate.candidate_id, approval)
```
建议 HTTP Action adapter:
### DeploymentPlan
```text
POST /api/silent-release/candidates
GET /api/silent-release/candidates/{candidate_id}
POST /api/silent-release/candidates/{candidate_id}/verify
POST /api/silent-release/candidates/{candidate_id}/approve-runtime
POST /api/silent-release/candidates/{candidate_id}/deploy
GET /api/silent-release/candidates/{candidate_id}/acceptance
service
release_path / release_root
dropin_path / dropin_sha256
rollback_path / rollback_sha256
source_commit
artifact_sha256
manifest_sha256
policy_sha256
health_contract
sha256
```
Tag interface 必须位于独立 namespace:
### SignedReleasePlanApproval
```text
POST /api/tag-releases/{version}/approve
POST /api/tag-releases/{version}/execute
GET /api/tag-releases/{version}/verify
release_plan_id
candidate_id
source_commit
tree_oid
artifact_sha256
manifest_sha256
deployment_plan_sha256
policy_sha256
approval_plan_id
evidence_closure_oid
owner_approval_record_oid
capabilities
expires_at
nonce
key_id
signature
fingerprint
```
runtime approval 不得被 TagController 接受;tag approval 不得被
DeploymentController 接受。
签名内容使用 canonical JSON,`signature` 不参与被签名字段。验证包括 Ed25519 签名、过期时间、
公钥状态、approval revoke、candidate/plan/policy 逐字段绑定和 capability allowlist。
## 6. ReleaseAcceptanceEngine
`CommandReleaseProbe` 提供实际只读 adapter:
- local main:指定 local repository;
- bare main:指定 authoritative bare Git;
- Gitea main/identity:认证 API,只从 credential reference 读取 token;
- artifact/manifest:流式 SHA-256 和 JSON 绑定;
- tests/canary:无 shell 的 argv command;
- runtime without `.git`;
- rollback artifact;
- authoritative Git 路径与可读性。
任何命令、Git、Gitea 或证据异常均转为 `FAILED`,不会降级为空成功。
## 7. ApprovalEnvelope
Owner 的 runtime envelope 只授权:
## 5. API
```text
action=production_runtime_switch
POST /createReleaseCandidate
GET /getReleaseCandidate?candidate_id=...
GET /getApprovalEnvelope?candidate_id=...
POST /submitOwnerApproval
POST /startDeployment
GET /getReleaseSaga?candidate_id=...
```
Envelope 绑定:
写接口要求 Bearer scope。`AIOS_SILENT_RELEASE_EXECUTION_ENABLED=false` 是默认值;此时可以
持久化 candidate、生成 approval envelope、保存经验证的 Owner approval 和读取 Saga,
但 `/startDeployment` 只返回 `SHADOW_MODE` 与空 `side_effects`。
- candidate_id/fingerprint;
- source commit/tree;
- artifact/manifest SHA;
- owner、approval ID、签发和过期时间。
## 6. PreSwitchValidator
`ApprovalLedger` 使用 `O_EXCL` 原子消费 approval ID,重放直接拒绝。
切换前重新收集并比较:
生产 tag 使用完全独立的 `TagApprovalEnvelope`:
- source commit 与 tree;
- artifact、manifest、policy 和 deployment plan SHA-256;
- rollback readiness;
- local、authoritative bare 与 Gitea refs;
- repository identity;
- task writer/production writer 边界;
- health contract。
任一变化都抛出:
```text
action=production_tag
version
tag_oid
target_oid
APPROVAL_STALE_REAPPROVAL_REQUIRED
```
## 8. DeploymentController
旧批准不会被“最佳努力”继续使用。
只接受 `READY_FOR_APPROVAL` 候选和完整 runtime envelope:
## 7. systemd 防线
1. 验证 Policy as Code;
2. 原子消费 approval;
3. 原子切换 versioned drop-in symlink;
4. 仅 restart 指定 GitOps runtime;
5. 验证 runtime commit 和 writer=false;
6. 任一异常切回旧 symlink并再次 restart;
7. rollback 后重新执行健康检查并确认 writer=false;
8. rollback 或 rollback health 失败时进入 `FAILED`,禁止伪装成功。
- service 与 drop-in 均为显式 allowlist;
- release 与 drop-in resolved path 必须位于 release root 内;
- 拒绝普通文件被替换为 symlink,拒绝 symlink escape;
- deployment lock 覆盖 switch、restart 和首次 health check;
- restart 前后记录 MainPID 与 ActiveEnterTimestampMonotonic;
- artifact、drop-in、rollback 和 manifest 均按 plan hash 复验;
- 子进程只接收固定 allowlist 环境,不继承 token、password、credential、askpass、
auth socket、`LD_PRELOAD`、`PYTHONPATH` 或数据库连接串。
Deployment adapter 的 interface 中没有 tag、Baseline 或 Registry 方法,因此 runtime
部署路径无法顺带调用这些能力。
## 8. TagController 分离
## 9. TagController 分离
TagController 必须独立完成 approval、execute、verify,并留下:
TagController 的三个动作严格分开:
```text
evidence_closure_oid
approval_plan_id
repository_identity
remote_tag_oid
remote_peeled_target_oid
gitea_api_tag_oid
gitea_listing_visible
```
1. `approve(envelope)`
2. `execute(approved)`
3. `verify(approved)`
同名 tag 的 tag OID 或 peeled target 不同,进入
`MANUAL_OWNER_REVIEW_REQUIRED`;自动流程不 push、不删除、不重写、不 force push。
同名 tag 已存在且 tag OID 或 target OID 不一致时抛出
`DuplicateTagConflict`。接口没有 delete、rewrite 或 force push。
## 9. Baseline / Registry 门禁
只有以下全部一致时,`baseline_registry_finalize` capability 才能完成:
```text
Registry = Baseline = remote tag = Gitea API = expected target
```
UNKNOWN、API 不可达、repository identity 不符或 listing 不可见均 fail closed。
## 10. Policy as Code
`config/gitops/silent-release-policy.json`:
```text
automatic: build, test, artifact, canary, evidence
owner: production_runtime_switch, production_tag
forbidden: force_push, tag_rewrite, direct_main_push,
baseline_mutation, registry_mutation
automatic:
build, test, artifact, canary, evidence
owner:
signed_release_plan
capabilities:
runtime_switch
tag_publish
baseline_registry_finalize
forbidden:
force_push
tag_rewrite
direct_main_push
baseline_mutation
registry_mutation
```
Policy 缺少任一部署禁止项时,DeploymentController 拒绝运行。
最后两个 forbidden 项指“脱离受治理 finalize controller 的直接 mutation”。
## 11. Security model
- fail-closed Git/Gitea/manifest/health adapter;
- credential 只从受控文件读取,不进入模型或持久化候选;
- candidate store 对 Authorization、token、credential、password、secret 字段脱敏;
- 错误文本统一清除 Authorization、token、URL userinfo 和 password;
- approval ID 一次性消费并有过期时间;
- candidate fingerprint 防止对其他 artifact 重放批准;
- runtime 默认 writer=false;
- systemd adapter 只允许管理 symlink,拒绝覆盖普通配置文件;
- no tag rewrite、force push、direct main push;
- runtime deployment 无 Baseline/Registry/tag capability。
- Ed25519 公钥 ring 支持 rotation 和 revoke,私钥不进入 runtime;
- approval plan 可撤销,过期、未知 key、错误签名均 fail closed;
- nonce 与 capability operation 持久化,重复相同请求幂等,不同 fingerprint 重放拒绝;
- Saga snapshot 原子写、journal append-only、CAS 防并发 writer;
- credential 只通过受控 reference 使用,不进入 Git、API 响应或异常;
- 子进程环境从空集合构建,不复制 `os.environ`;
- Shadow 模式默认开启,execution adapter 缺失时不能降级执行;
- runtime、tag、Baseline/Registry 分属三个能力最小化控制器。
## 12. Migration plan
### Phase 1:shadow
1. Shadow:生成 candidate、plan、acceptance、approval package,仅比较现有人工结果。
2. Signed approval:接入 Owner 公钥 ring、撤销源和审计 record;仍不接生产 executor。
3. Canary execution:使用非生产 service/release root 验证锁、PID、health、崩溃恢复和回滚。
4. Controlled production switch:另行 Owner 授权启用 execution flag,只开放 runtime capability。
5. Tag/finalize:分别完成 Gitea 与 Registry/Baseline 生产验收后再启用对应 capability。
- 只生成 ReleaseCandidate 和 acceptance evidence;
- 与现有人工报告并行比较;
- 不接入 production executor。
### Phase 2:approval envelope
- Owner UI/Action 只生成一次性 runtime envelope;
- production 仍由现有受控 executor 执行;
- 验证 approval replay 和过期处理。
### Phase 3:controlled deployment
- 接入 versioned drop-in symlink;
- 先在 canary service 完成自动 switch/restart/rollback;
- production writer 始终为 false。
### Phase 4:silent release
- 自动 build/test/artifact/canary/evidence;
- READY_FOR_APPROVAL 时只通知一次 Owner;
- 获得 runtime envelope 后自动部署和验收;
- TagController 始终保持独立授权。
每个阶段均可退回上一个阶段;不开启下一能力不会影响已经启用的控制器。
## 13. Rollback plan
- 切换前记录旧 symlink target;
- 新 runtime 任一健康、Git、hash、writer 或一致性门禁失败即恢复旧 target;
- 只 restart GitOps runtime;
- rollback 后再次验证旧 runtime、writer=false 和只读 Git-backed API;
- 保留失败候选、脱敏证据和 approval ledger;
- 不使用 Git reset、force push、tag rewrite;
- rollback 不修改 Baseline、Registry 或 production tag。
- switch 前保存旧 symlink target、PID、启动时间、artifact 与 plan fingerprint;
- restart 或 health 任一失败进入 `ROLLBACK_PENDING`;
- 原子恢复旧 target,只 restart 指定 allowlist service;
- 重新验证旧 runtime commit、writer=false 与 repository identity;
- 成功进入 `ROLLED_BACK`,失败进入 `MANUAL_OWNER_REVIEW_REQUIRED`;
- rollback 不修改 Git main、production tag、Baseline 或 Registry;
- 保留脱敏 journal、approval fingerprint 和失败证据。
## 14. 自动测试矩阵
## 14. 测试与外部依赖边界
实现测试覆盖:
自动测试覆盖签名伪造、字段篡改、过期/revoke/key rotation、nonce replay、CAS race、
崩溃恢复、artifact/policy/plan mutation、tag conflict、secret leakage、runtime without
`.git`、rollback 和 shadow no-side-effects。
- artifact mismatch;
- Gitea unavailable;
- authoritative Git unavailable;
- health failure rollback;
- approval replay;
- duplicate tag conflict;
- secret leakage;
- runtime without `.git`;
- state transition;
- policy separation;
- candidate persistence;
- automatic prepare 到单次 Owner gate。
本隔离实现不创建生产 KMS/key、不安装 systemd unit、不部署 Action runtime、不创建 tag,
也不修改 Baseline/Registry。上述动作必须在单独生产批准中执行。

Some files were not shown because too many files have changed in this diff Show More