feat(gitops): harden one-approval silent release saga

active-runtime/app/aios_chatgpt_actions_api.py

@@ -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 = [
@@ -80,11 +100,19 @@ READ_ONLY_GET_PATHS = frozenset({
     '/admin/ai-evolution/api/data',
     '/api/release-consistency', '/api/tag-publication-status',
     '/api/release-acceptance',
+    '/getReleaseCandidate', '/getApprovalEnvelope', '/getReleaseSaga',
+})
+SILENT_RELEASE_POST_PATHS = frozenset({
+    '/createReleaseCandidate',
+    '/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 +129,32 @@ 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 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,
+    )
 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 +522,36 @@ 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',
+                '/getApprovalEnvelope',
+                '/getReleaseSaga',
+            ):
+                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)
+                    )
+                return self.sendj(service.get_release_saga(candidate_id))
                         # 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 +567,32 @@ 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 == '/submitOwnerApproval':
+                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
+                approval=(
+                    SignedReleasePlanApproval.from_dict(data['approval'])
+                    if isinstance(data.get('approval'), dict)
+                    else None
+                )
+                return self.sendj(
+                    silent_release_service().start_deployment(
+                        str(data.get('candidate_id') or ''),
+                        approval=approval,
+                    )
+                )
             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 +603,22 @@ 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 +820,7 @@ 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}')

active-runtime/app/silent_release_controller.py

@@ -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:

config/gitops/silent-release-policy.json

@@ -1,5 +1,5 @@
 {
-  "schema_version": 1,
+  "schema_version": 2,
   "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 +27,37 @@
     "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",
+    "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
   }
 }

docs/gitops/SILENT_RELEASE_CONTROLLER_DESIGN.md

@@ -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。

docs/gitops/openapi-chatgpt-actions.yaml

@@ -135,7 +135,142 @@ components:
       properties:
         text: {type: string}
         branch: {type: string}
+    DeploymentPlan:
+      type: object
+      required: [service, release_path, release_root, dropin_path, dropin_sha256, rollback_path, rollback_sha256, source_commit, artifact_sha256, manifest_sha256, policy_sha256, health_contract]
+      properties:
+        service: {type: string}
+        release_path: {type: string}
+        release_root: {type: string}
+        dropin_path: {type: string}
+        dropin_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        rollback_path: {type: string}
+        rollback_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        source_commit: {type: string, pattern: '^[0-9a-f]{40}$'}
+        artifact_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        manifest_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        policy_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        health_contract: {type: object, additionalProperties: true}
+    ReleaseCandidateInput:
+      type: object
+      required: [candidate_id, source_commit, tree_oid, artifact_sha256, manifest_sha256]
+      properties:
+        candidate_id: {type: string, pattern: '^[A-Za-z0-9][A-Za-z0-9_.-]{2,127}$'}
+        source_commit: {type: string, pattern: '^[0-9a-f]{40}$'}
+        tree_oid: {type: string, pattern: '^[0-9a-f]{40}$'}
+        artifact_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        manifest_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        tests: {type: object}
+        canary: {type: object}
+        rollback_ready: {type: boolean}
+        git_consistency: {type: object}
+        gitea_consistency: {type: object}
+        deployment_plan: {$ref: '#/components/schemas/DeploymentPlan'}
+    SignedReleasePlanApproval:
+      type: object
+      required: [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]
+      properties:
+        release_plan_id: {type: string}
+        candidate_id: {type: string}
+        source_commit: {type: string, pattern: '^[0-9a-f]{40}$'}
+        tree_oid: {type: string, pattern: '^[0-9a-f]{40}$'}
+        artifact_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        manifest_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        deployment_plan_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        policy_sha256: {type: string, pattern: '^[0-9a-f]{64}$'}
+        approval_plan_id: {type: string}
+        evidence_closure_oid: {type: string, pattern: '^[0-9a-f]{40}$'}
+        owner_approval_record_oid: {type: string, pattern: '^[0-9a-f]{40}$'}
+        capabilities:
+          type: array
+          uniqueItems: true
+          items:
+            type: string
+            enum: [runtime_switch, tag_publish, baseline_registry_finalize]
+        expires_at: {type: string, format: date-time}
+        nonce: {type: string, minLength: 16}
+        key_id: {type: string}
+        signature: {type: string, format: byte, writeOnly: true}
+    StartDeploymentRequest:
+      type: object
+      required: [candidate_id, approval]
+      properties:
+        candidate_id: {type: string}
+        approval: {$ref: '#/components/schemas/SignedReleasePlanApproval'}
 paths:
+  /createReleaseCandidate:
+    post:
+      operationId: createReleaseCandidate
+      summary: Persist an immutable release candidate and deployment plan; shadow-safe
+      security: [{bearerAuth: []}]
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema: {$ref: '#/components/schemas/ReleaseCandidateInput'}
+      responses:
+        '200': {description: Candidate created or idempotently returned}
+        '400': {description: Invalid or conflicting candidate evidence}
+        '401': {description: Unauthorized}
+      x-openai-isConsequential: true
+  /getReleaseCandidate:
+    get:
+      operationId: getReleaseCandidate
+      summary: Read a durable release candidate
+      parameters:
+        - {in: query, name: candidate_id, required: true, schema: {type: string}}
+      responses:
+        '200': {description: Candidate}
+        '404': {description: Candidate not found}
+  /getApprovalEnvelope:
+    get:
+      operationId: getApprovalEnvelope
+      summary: Generate the immutable owner-signature payload
+      parameters:
+        - {in: query, name: candidate_id, required: true, schema: {type: string}}
+      responses:
+        '200': {description: Approval envelope fingerprint inputs}
+        '404': {description: Candidate not found}
+  /submitOwnerApproval:
+    post:
+      operationId: submitOwnerApproval
+      summary: Verify and durably store a signed owner release plan
+      security: [{bearerAuth: []}]
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema: {$ref: '#/components/schemas/SignedReleasePlanApproval'}
+      responses:
+        '200': {description: Signature verified and approval stored}
+        '400': {description: Invalid approval}
+        '403': {description: Signature, key, expiry, or revocation failure}
+        '409': {description: Conflicting release plan}
+      x-openai-isConsequential: true
+  /startDeployment:
+    post:
+      operationId: startDeployment
+      summary: Start the approved Saga; default shadow mode returns no side effects
+      security: [{bearerAuth: []}]
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema: {$ref: '#/components/schemas/StartDeploymentRequest'}
+      responses:
+        '200': {description: Shadow result or deployment Saga result}
+        '403': {description: Execution disabled or approval invalid}
+        '409': {description: Saga conflict}
+      x-openai-isConsequential: true
+  /getReleaseSaga:
+    get:
+      operationId: getReleaseSaga
+      summary: Read durable Saga state and event journal
+      parameters:
+        - {in: query, name: candidate_id, required: true, schema: {type: string}}
+      responses:
+        '200': {description: Saga snapshot}
+        '404': {description: Saga not found}
   /api/base:
     get:
       operationId: getBaseline

tests/test_silent_release_controller.py

@@ -154,6 +154,7 @@ class SilentReleaseControllerTests(unittest.TestCase):
                     ApprovalLedger(Path(td) / "approvals"),
                     SilentReleasePolicy.default(),
                     store,
+                    allow_legacy_unsigned_approval=True,
                 ),
                 store=store,
                 policy=SilentReleasePolicy.default(),
@@ -248,6 +249,7 @@ class SilentReleaseControllerTests(unittest.TestCase):
                 deployment,
                 ApprovalLedger(Path(td) / "approvals"),
                 SilentReleasePolicy.default(),
+                allow_legacy_unsigned_approval=True,
             )
             result = controller.deploy(value, envelope)
             self.assertEqual(value.state, ReleaseState.VERIFIED)
@@ -273,6 +275,7 @@ class SilentReleaseControllerTests(unittest.TestCase):
                 deployment,
                 ApprovalLedger(Path(td) / "approvals"),
                 SilentReleasePolicy.default(),
+                allow_legacy_unsigned_approval=True,
             ).deploy(value, envelope)
             self.assertFalse(result["verified"])
             self.assertEqual(value.state, ReleaseState.ROLLED_BACK)
@@ -292,7 +295,10 @@ class SilentReleaseControllerTests(unittest.TestCase):
         with tempfile.TemporaryDirectory() as td:
             ledger = ApprovalLedger(Path(td) / "approvals")
             controller = DeploymentController(
-                FakeDeployment(), ledger, SilentReleasePolicy.default()
+                FakeDeployment(),
+                ledger,
+                SilentReleasePolicy.default(),
+                allow_legacy_unsigned_approval=True,
             )
             controller.deploy(value, envelope)
             second = candidate()
@@ -309,6 +315,7 @@ class SilentReleaseControllerTests(unittest.TestCase):
             adapter,
             SilentReleasePolicy.default(),
             ledger=None,
+            allow_legacy_unsigned_approval=True,
         )
         envelope = TagApprovalEnvelope(
             approval_id="tag-approval-001",