feat(governance): 新增 post-release verifier rollback gate
Some checks failed
Code Review / ai-code-review (push) Successful in 14s
CD Pipeline / tests (push) Successful in 1m30s
CD Pipeline / post-deploy-checks (push) Has been cancelled
CD Pipeline / build-and-deploy (push) Has been cancelled

This commit is contained in:
Your Name
2026-06-14 05:33:48 +08:00
parent 33dc2f416c
commit 040c320c5e
13 changed files with 1676 additions and 6 deletions

View File

@@ -142,6 +142,9 @@ from src.services.ai_agent_result_capture_owner_approved_release_readiness_readb
from src.services.ai_agent_result_capture_owner_release_approval_gate import (
load_latest_ai_agent_result_capture_owner_release_approval_gate,
)
from src.services.ai_agent_result_capture_post_release_verifier_rollback_gate import (
load_latest_ai_agent_result_capture_post_release_verifier_rollback_gate,
)
from src.services.ai_agent_result_capture_promotion_approval_gate import (
load_latest_ai_agent_result_capture_promotion_approval_gate,
)
@@ -2051,6 +2054,37 @@ async def get_agent_result_capture_owner_release_approval_gate() -> dict[str, An
) from exc
@router.get(
"/agent-result-capture-post-release-verifier-rollback-gate",
response_model=dict[str, Any],
summary="取得 AI Agent result capture post-release verifier rollback gate",
description=(
"讀取最新已提交的 P2-132 post-release verifier / rollback gate"
"此端點只回傳 post-release verifier gate、rollback release gate、release verification hold、"
"live apply post-release gate、blocked post-release transition 與 operator handoff"
"不批准 owner release、不批准 maintenance window、不確認 rollback owner、不釋放 live apply、"
"不套用 writer、不寫 receipt、不寫 result capture、learning、PlayBook trust、reviewer queue、"
"Gateway queue不送 Telegram、不呼叫 Bot API、不讀 secret。"
),
)
async def get_agent_result_capture_post_release_verifier_rollback_gate() -> dict[str, Any]:
"""Return the latest read-only post-release verifier / rollback gate."""
try:
payload = await asyncio.to_thread(load_latest_ai_agent_result_capture_post_release_verifier_rollback_gate)
return redact_public_lan_topology(payload)
except FileNotFoundError as exc:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail=str(exc),
) from exc
except (json.JSONDecodeError, ValueError) as exc:
logger.error("ai_agent_result_capture_post_release_verifier_rollback_gate_invalid", error=str(exc))
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail="AI Agent result capture post-release verifier rollback gate 無效",
) from exc
@router.get(
"/agent-owner-approved-fixture-dry-run",
response_model=dict[str, Any],

View File

@@ -0,0 +1,408 @@
"""
AI Agent result capture post-release verifier rollback gate snapshot.
Loads the latest committed P2-132 post-release verifier / rollback gate.
This module validates committed evidence only; it never approves owner release,
approves maintenance windows, confirms rollback owners, releases live apply,
applies writers, writes receipts, writes result captures, writes learning
records, updates PlayBook trust, writes reviewer / Gateway queues, sends
Telegram messages, reads secrets, or performs destructive operations.
"""
from __future__ import annotations
import json
from pathlib import Path
from typing import Any
from src.services.snapshot_paths import default_evaluations_dir
_DEFAULT_EVALUATIONS_DIR = default_evaluations_dir(Path(__file__))
_SNAPSHOT_PATTERN = "ai_agent_result_capture_post_release_verifier_rollback_gate_*.json"
_SCHEMA_VERSION = "ai_agent_result_capture_post_release_verifier_rollback_gate_v1"
_RUNTIME_AUTHORITY = "result_capture_post_release_verifier_rollback_gate_only_no_live_write"
def load_latest_ai_agent_result_capture_post_release_verifier_rollback_gate(
evaluations_dir: Path | None = None,
) -> dict[str, Any]:
"""Load the newest committed post-release verifier / rollback gate."""
directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
if not candidates:
raise FileNotFoundError(f"no AI Agent result capture post-release verifier rollback gate snapshots found in {directory}")
latest = candidates[-1]
with latest.open(encoding="utf-8") as handle:
payload = json.load(handle)
if not isinstance(payload, dict):
raise ValueError(f"{latest}: expected JSON object")
label = str(latest)
_require_schema(payload, label)
_require_prior(payload, label)
_require_truth(payload, label)
_require_verifier_gates(payload, label)
_require_rollback_gates(payload, label)
_require_verification_holds(payload, label)
_require_live_apply_gates(payload, label)
_require_blocked_transitions(payload, label)
_require_actions(payload, label)
_require_display_redaction(payload, label)
_require_no_forbidden_display_terms(payload, label)
_require_rollup_consistency(payload, label)
return payload
def _require_schema(payload: dict[str, Any], label: str) -> None:
if payload.get("schema_version") != _SCHEMA_VERSION:
raise ValueError(f"{label}: expected schema_version={_SCHEMA_VERSION}")
status = payload.get("program_status") or {}
expected = {
"current_priority": "P2",
"current_task_id": "P2-132",
"next_task_id": "P2-133",
"read_only_mode": True,
"runtime_authority": _RUNTIME_AUTHORITY,
"overall_completion_percent": 100,
}
mismatches = _mismatches(status, expected)
if mismatches:
raise ValueError(f"{label}: program_status mismatch: {mismatches}")
if not status.get("status_note"):
raise ValueError(f"{label}: program_status.status_note is required")
def _require_prior(payload: dict[str, Any], label: str) -> None:
prior = payload.get("prior_owner_release_approval_gate") or {}
expected = {
"schema_version": "ai_agent_result_capture_owner_release_approval_gate_v1",
"owner_release_approval_packet_count": 5,
"maintenance_window_approval_gate_count": 5,
"live_apply_release_approval_gate_count": 5,
"rollback_owner_approval_check_count": 5,
"blocked_approval_transition_count": 6,
"operator_action_count": 5,
"approval_required_total": 8,
"blocked_total": 9,
"owner_release_approved_count": 0,
"maintenance_window_approved_count": 0,
"rollback_owner_confirmed_count": 0,
"post_release_verifier_ready_count": 0,
"release_approval_pass_count": 0,
"live_apply_release_pass_count": 0,
"writer_apply_count": 0,
"execution_apply_count": 0,
"receipt_write_count": 0,
"result_capture_write_count": 0,
"learning_write_count": 0,
"playbook_trust_write_count": 0,
"reviewer_queue_write_count": 0,
"gateway_queue_write_count": 0,
"telegram_send_count": 0,
"bot_api_call_count": 0,
"report_receipt_write_count": 0,
}
mismatches = _mismatches(prior, expected)
if mismatches:
raise ValueError(f"{label}: prior_owner_release_approval_gate mismatch: {mismatches}")
if not prior.get("readiness_note"):
raise ValueError(f"{label}: prior_owner_release_approval_gate.readiness_note is required")
def _require_truth(payload: dict[str, Any], label: str) -> None:
truth = payload.get("post_release_verifier_truth") or {}
required_true = {
"p2_131_approval_gate_loaded",
"post_release_verifier_gate_ready",
"rollback_release_gate_ready",
"release_verification_hold_active",
"live_apply_post_release_hold_active",
"owner_release_review_still_required",
"maintenance_window_review_still_required",
"rollback_owner_review_required",
"redaction_review_required",
"post_release_verifier_gate_only",
}
missing = sorted(field for field in required_true if truth.get(field) is not True)
if missing:
raise ValueError(f"{label}: post-release verifier gate flags must remain true: {missing}")
required_false = {
"owner_release_approved",
"maintenance_window_approved",
"rollback_owner_confirmed",
"post_release_verifier_ready",
"release_verification_passed",
"rollback_release_passed",
"live_apply_release_passed",
"writer_apply_enabled",
"execution_apply_enabled",
"receipt_write_enabled",
"reviewer_queue_write_enabled",
"gateway_queue_write_enabled",
"telegram_send_enabled",
"bot_api_call_enabled",
"report_receipt_write_enabled",
"result_capture_write_enabled",
"learning_write_enabled",
"playbook_trust_write_enabled",
"production_write_enabled",
"secret_read_enabled",
"destructive_operation_enabled",
}
unsafe = sorted(field for field in required_false if truth.get(field) is not False)
if unsafe:
raise ValueError(f"{label}: post-release verifier live/send/write flags must remain false: {unsafe}")
zero_counts = {
"owner_release_approved_count",
"maintenance_window_approved_count",
"rollback_owner_confirmed_count",
"post_release_verifier_ready_count",
"release_verification_pass_count",
"rollback_release_pass_count",
"live_apply_release_pass_count",
"writer_apply_count",
"execution_apply_count",
"receipt_write_count",
"reviewer_queue_write_count",
"gateway_queue_write_count",
"telegram_send_count",
"bot_api_call_count",
"report_receipt_write_count",
"result_capture_write_count",
"learning_write_count",
"playbook_trust_write_count",
"production_write_count",
"secret_read_count",
"destructive_operation_count",
}
non_zero = sorted(field for field in zero_counts if truth.get(field) != 0)
if non_zero:
raise ValueError(f"{label}: post-release verifier live counters must remain zero: {non_zero}")
if not truth.get("truth_note"):
raise ValueError(f"{label}: post_release_verifier_truth.truth_note is required")
def _require_verifier_gates(payload: dict[str, Any], label: str) -> None:
items = payload.get("post_release_verifier_gates") or []
_require_ids(
items,
"gate_id",
{
"post_release_verifier_result_capture",
"post_release_verifier_learning",
"post_release_verifier_playbook_trust",
"post_release_verifier_reviewer_queue",
"post_release_verifier_gateway_queue",
},
label,
"post-release verifier gates",
)
for item in items:
item_id = item.get("gate_id")
if item.get("gate_mode") != "post_release_verifier_gate":
raise ValueError(f"{label}: post-release verifier gate {item_id} mode is invalid")
if item.get("post_release_verifier_ready") is not False or item.get("release_verification_passed") is not False:
raise ValueError(f"{label}: post-release verifier gate {item_id} must stay unready and unpassed")
if item.get("status") not in {"ready_for_owner_review", "approval_required", "blocked_by_policy"}:
raise ValueError(f"{label}: post-release verifier gate {item_id} status is invalid")
if not item.get("verifier_summary") or not _is_redacted_sha256(item.get("evidence_hash")):
raise ValueError(f"{label}: post-release verifier gate {item_id} must include summary and redacted evidence_hash")
def _require_rollback_gates(payload: dict[str, Any], label: str) -> None:
items = payload.get("rollback_release_gates") or []
_require_count(items, 5, label, "rollback release gates")
for item in items:
item_id = item.get("gate_id")
if item.get("gate_mode") != "rollback_release_gate":
raise ValueError(f"{label}: rollback release gate {item_id} mode is invalid")
if item.get("rollback_owner_required") is not True:
raise ValueError(f"{label}: rollback release gate {item_id} must require rollback owner")
if item.get("rollback_owner_confirmed") is not False or item.get("rollback_release_enabled") is not False:
raise ValueError(f"{label}: rollback release gate {item_id} must stay unconfirmed and disabled")
if item.get("status") not in {"ready_for_owner_review", "approval_required", "blocked_by_policy"}:
raise ValueError(f"{label}: rollback release gate {item_id} status is invalid")
def _require_verification_holds(payload: dict[str, Any], label: str) -> None:
items = payload.get("release_verification_holds") or []
_require_count(items, 5, label, "release verification holds")
for item in items:
item_id = item.get("hold_id")
if item.get("hold_mode") != "release_verification_hold":
raise ValueError(f"{label}: release verification hold {item_id} mode is invalid")
if item.get("owner_release_approved") is not False or item.get("maintenance_window_approved") is not False:
raise ValueError(f"{label}: release verification hold {item_id} must stay unapproved")
if item.get("release_verification_passed") is not False:
raise ValueError(f"{label}: release verification hold {item_id} must stay unpassed")
if item.get("status") not in {"ready_for_owner_review", "approval_required", "blocked_by_policy"}:
raise ValueError(f"{label}: release verification hold {item_id} status is invalid")
def _require_live_apply_gates(payload: dict[str, Any], label: str) -> None:
items = payload.get("live_apply_post_release_gates") or []
_require_count(items, 5, label, "live apply post-release gates")
for item in items:
item_id = item.get("gate_id")
if item.get("gate_mode") != "live_apply_post_release_gate":
raise ValueError(f"{label}: live apply post-release gate {item_id} mode is invalid")
if item.get("post_release_verifier_ready") is not False or item.get("live_apply_release_enabled") is not False:
raise ValueError(f"{label}: live apply post-release gate {item_id} must stay held and disabled")
if item.get("status") not in {"ready_for_owner_review", "approval_required", "blocked_by_policy"}:
raise ValueError(f"{label}: live apply post-release gate {item_id} status is invalid")
def _require_blocked_transitions(payload: dict[str, Any], label: str) -> None:
items = payload.get("blocked_post_release_transitions") or []
_require_count(items, 6, label, "blocked post-release transitions")
critical = 0
for item in items:
item_id = item.get("blocker_id")
if item.get("status") not in {"approval_required", "blocked_by_policy"}:
raise ValueError(f"{label}: blocked post-release transition {item_id} status is invalid")
if item.get("severity") == "critical":
critical += 1
if not item.get("blocked_action") or not _is_redacted_sha256(item.get("evidence_hash")):
raise ValueError(f"{label}: blocked post-release transition {item_id} must include action and redacted evidence_hash")
if critical != 5:
raise ValueError(f"{label}: expected 5 critical blocked post-release transitions, got {critical}")
def _require_actions(payload: dict[str, Any], label: str) -> None:
items = payload.get("operator_actions") or []
_require_ids(
items,
"action_id",
{
"review_post_release_verifier_gates",
"verify_rollback_release_gate",
"verify_release_verification_hold",
"verify_live_apply_post_release_hold",
"prepare_p2_133_final_release_candidate_readback",
},
label,
"operator actions",
)
for item in items:
if item.get("runtime_write_allowed") is not False:
raise ValueError(f"{label}: operator action {item.get('action_id')} must keep runtime_write_allowed=false")
if item.get("status") not in {"ready_for_operator_review", "approval_required", "blocked_by_policy"}:
raise ValueError(f"{label}: operator action {item.get('action_id')} status is invalid")
def _require_display_redaction(payload: dict[str, Any], label: str) -> None:
contract = payload.get("display_redaction_contract") or {}
expected = {
"redaction_required": True,
"raw_prompt_display_allowed": False,
"private_reasoning_display_allowed": False,
"secret_value_display_allowed": False,
"raw_runtime_payload_display_allowed": False,
"internal_collaboration_content_display_allowed": False,
}
mismatches = _mismatches(contract, expected)
if mismatches:
raise ValueError(f"{label}: display_redaction_contract mismatch: {mismatches}")
if not contract.get("frontend_display_policy"):
raise ValueError(f"{label}: display_redaction_contract.frontend_display_policy is required")
def _require_no_forbidden_display_terms(payload: dict[str, Any], label: str) -> None:
forbidden = {
"work_window_transcript",
"session_id",
"browser_context",
"authorization_header",
"raw Telegram payload",
"private reasoning",
"raw prompt",
"chain-of-thought",
"批准!繼續",
"My request for Codex",
"In app browser",
}
text = json.dumps(payload, ensure_ascii=False)
present = sorted(term for term in forbidden if term in text)
if present:
raise ValueError(f"{label}: forbidden display terms present: {present}")
def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
rollups = payload.get("rollups") or {}
expected = {
"post_release_verifier_gate_count": len(payload.get("post_release_verifier_gates") or []),
"rollback_release_gate_count": len(payload.get("rollback_release_gates") or []),
"release_verification_hold_count": len(payload.get("release_verification_holds") or []),
"live_apply_post_release_gate_count": len(payload.get("live_apply_post_release_gates") or []),
"blocked_post_release_transition_count": len(payload.get("blocked_post_release_transitions") or []),
"operator_action_count": len(payload.get("operator_actions") or []),
"approval_required_verifier_count": _count_status(payload.get("post_release_verifier_gates"), "approval_required"),
"blocked_verifier_count": _count_status(payload.get("post_release_verifier_gates"), "blocked_by_policy"),
"approval_required_rollback_count": _count_status(payload.get("rollback_release_gates"), "approval_required"),
"blocked_rollback_count": _count_status(payload.get("rollback_release_gates"), "blocked_by_policy"),
"approval_required_verification_hold_count": _count_status(payload.get("release_verification_holds"), "approval_required"),
"blocked_verification_hold_count": _count_status(payload.get("release_verification_holds"), "blocked_by_policy"),
"approval_required_live_apply_count": _count_status(payload.get("live_apply_post_release_gates"), "approval_required"),
"blocked_live_apply_count": _count_status(payload.get("live_apply_post_release_gates"), "blocked_by_policy"),
"critical_blocker_count": sum(1 for item in payload.get("blocked_post_release_transitions") or [] if item.get("severity") == "critical"),
}
zero_fields = {
"owner_release_approved_count",
"maintenance_window_approved_count",
"rollback_owner_confirmed_count",
"post_release_verifier_ready_count",
"release_verification_pass_count",
"rollback_release_pass_count",
"live_apply_release_pass_count",
"writer_apply_count",
"execution_apply_count",
"receipt_write_count",
"reviewer_queue_write_count",
"gateway_queue_write_count",
"telegram_send_count",
"bot_api_call_count",
"report_receipt_write_count",
"result_capture_write_count",
"learning_write_count",
"playbook_trust_write_count",
"production_write_count",
"secret_read_count",
"destructive_operation_count",
}
expected.update({field: 0 for field in zero_fields})
mismatches = _mismatches(rollups, expected)
if mismatches:
raise ValueError(f"{label}: rollup counts mismatch: {mismatches}")
def _mismatches(source: dict[str, Any], expected: dict[str, Any]) -> dict[str, dict[str, Any]]:
return {key: {"expected": value, "actual": source.get(key)} for key, value in expected.items() if source.get(key) != value}
def _require_count(items: Any, expected_count: int, label: str, name: str) -> None:
if not isinstance(items, list) or len(items) != expected_count:
raise ValueError(f"{label}: expected {expected_count} {name}, got {len(items) if isinstance(items, list) else 'non-list'}")
def _require_ids(items: list[dict[str, Any]], id_field: str, expected_ids: set[str], label: str, name: str) -> None:
_require_count(items, len(expected_ids), label, name)
found = {str(item.get(id_field)) for item in items}
if found != expected_ids:
raise ValueError(f"{label}: {name} ids mismatch: expected={sorted(expected_ids)} actual={sorted(found)}")
def _count_status(items: Any, status: str) -> int:
if not isinstance(items, list):
return 0
return sum(1 for item in items if isinstance(item, dict) and item.get("status") == status)
def _is_redacted_sha256(value: Any) -> bool:
if not isinstance(value, str):
return False
prefix = "sha256:"
return value.startswith(prefix) and len(value) == len(prefix) + 64