feat(governance): 新增 runtime write gate review

apps/api/src/api/v1/agents.py

@@ -82,6 +82,9 @@ from src.services.ai_agent_proactive_operations_contract import (
 from src.services.ai_agent_redis_dry_run_gate import (
     load_latest_ai_agent_redis_dry_run_gate,
 )
+from src.services.ai_agent_runtime_write_gate_review import (
+    load_latest_ai_agent_runtime_write_gate_review,
+)
 from src.services.ai_agent_telegram_action_required_digest_policy import (
     load_latest_ai_agent_telegram_action_required_digest_policy,
 )
@@ -757,6 +760,34 @@ async def get_agent_owner_approved_learning_dry_run() -> dict[str, Any]:
         ) from exc
 
 
+@router.get(
+    "/agent-runtime-write-gate-review",
+    response_model=dict[str, Any],
+    summary="取得 AI Agent runtime write gate review",
+    description=(
+        "讀取最新已提交的 runtime write gate review 契約；此端點只回傳雙重批准、"
+        "dry-run hash、post-write verifier 與 redaction 欄位檢查，"
+        "不寫 KM、不更新 PlayBook trust、不寫 timeline、不寫 replay score、不發 Telegram、"
+        "不啟動 runtime worker、不回傳未脫敏 payload。"
+    ),
+)
+async def get_agent_runtime_write_gate_review() -> dict[str, Any]:
+    """Return the latest read-only AI Agent runtime write gate review."""
+    try:
+        return await asyncio.to_thread(load_latest_ai_agent_runtime_write_gate_review)
+    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_runtime_write_gate_review_invalid", error=str(exc))
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="AI Agent runtime write gate review 無效",
+        ) from exc
+
+
 @router.get(
     "/agent-owner-approved-fixture-dry-run",
     response_model=dict[str, Any],

apps/api/src/services/ai_agent_runtime_write_gate_review.py

@@ -0,0 +1,147 @@
+"""
+AI Agent runtime write gate review snapshot.
+
+Loads the latest committed P2-403G runtime write gate review contract. This
+module never writes KM, PlayBook trust, timeline learning, replay scores, or
+Telegram receipts.
+"""
+
+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_runtime_write_gate_review_*.json"
+_SCHEMA_VERSION = "ai_agent_runtime_write_gate_review_v1"
+
+
+def load_latest_ai_agent_runtime_write_gate_review(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed AI Agent runtime write gate review contract."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no AI Agent runtime write gate review 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")
+    _require_schema(payload, str(latest))
+    _require_write_boundaries(payload, str(latest))
+    _require_review_contract(payload, str(latest))
+    _require_rollup_consistency(payload, str(latest))
+    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 {}
+    if status.get("read_only_mode") is not True:
+        raise ValueError(f"{label}: program_status.read_only_mode must be true")
+    if status.get("runtime_authority") != "write_gate_review_only_no_runtime_write":
+        raise ValueError(f"{label}: runtime_authority must remain write_gate_review_only_no_runtime_write")
+
+
+def _require_write_boundaries(payload: dict[str, Any], label: str) -> None:
+    boundaries = payload.get("approval_boundaries") or {}
+    enabled = sorted(key for key, value in boundaries.items() if value is not False)
+    if enabled:
+        raise ValueError(f"{label}: approval boundaries must remain false: {enabled}")
+
+    truth = payload.get("runtime_write_truth") or {}
+    false_flags = {
+        "runtime_write_allowed",
+        "km_write_allowed",
+        "playbook_trust_write_allowed",
+        "timeline_learning_write_allowed",
+        "agent_replay_score_write_allowed",
+        "telegram_send_allowed",
+    }
+    unsafe = sorted(flag for flag in false_flags if truth.get(flag) is not False)
+    if unsafe:
+        raise ValueError(f"{label}: runtime write flags must remain false: {unsafe}")
+
+    required_true = {
+        "dual_approval_required",
+        "dry_run_hash_required",
+        "post_write_verifier_required",
+    }
+    missing = sorted(flag for flag in required_true if truth.get(flag) is not True)
+    if missing:
+        raise ValueError(f"{label}: required write gates must remain true: {missing}")
+
+    zero_counts = {
+        "dual_approval_received_count",
+        "dry_run_hash_verified_count",
+        "post_write_verifier_pass_count",
+    }
+    non_zero = sorted(key for key in zero_counts if truth.get(key) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: write gate counts must remain zero: {non_zero}")
+
+
+def _require_review_contract(payload: dict[str, Any], label: str) -> None:
+    review = payload.get("write_gate_review") or {}
+    required_fields = set(review.get("required_fields") or [])
+    required_minimum = {
+        "dual_approval_ids",
+        "dry_run_preview_hash",
+        "redacted_evidence_refs",
+        "target_write_surface",
+        "rollback_owner",
+        "post_write_verifier_ref",
+    }
+    missing = sorted(required_minimum - required_fields)
+    if missing:
+        raise ValueError(f"{label}: write gate review missing required fields: {missing}")
+
+    verification = payload.get("post_write_verification") or {}
+    if verification.get("verification_required") is not True:
+        raise ValueError(f"{label}: post-write verification must be required")
+    if verification.get("rollback_required") is not True:
+        raise ValueError(f"{label}: rollback must be required")
+    if not verification.get("verification_steps"):
+        raise ValueError(f"{label}: verification steps must not be empty")
+
+    redaction = payload.get("display_redaction_contract") or {}
+    if redaction.get("redaction_required") is not True:
+        raise ValueError(f"{label}: frontend redaction must be required")
+    for flag in ("raw_payload_display_allowed", "private_reasoning_display_allowed", "secret_value_display_allowed"):
+        if redaction.get(flag) is not False:
+            raise ValueError(f"{label}: {flag} must remain false")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    rollups = payload.get("rollups") or {}
+    targets = payload.get("write_targets") or []
+    gates = payload.get("approval_gates") or []
+    review = payload.get("write_gate_review") or {}
+    expected_counts = {
+        "write_target_count": len(targets),
+        "approval_gate_count": len(gates),
+        "blocked_runtime_action_count": len({gate.get("blocked_runtime_action") for gate in gates}),
+        "required_field_count": len(review.get("required_fields") or []),
+        "forbidden_field_count": len(review.get("forbidden_fields") or []),
+    }
+    mismatched = {
+        key: {"expected": expected, "actual": rollups.get(key)}
+        for key, expected in expected_counts.items()
+        if rollups.get(key) != expected
+    }
+    if mismatched:
+        raise ValueError(f"{label}: rollup counts must match payload sections: {mismatched}")
+
+    approval_required = sorted(gate.get("gate_id") for gate in gates if gate.get("status") == "approval_required")
+    if sorted(rollups.get("approval_required_gate_ids") or []) != approval_required:
+        raise ValueError(f"{label}: rollups.approval_required_gate_ids mismatch")
+    if rollups.get("live_write_count_total") != 0:
+        raise ValueError(f"{label}: live write count must remain zero")

apps/api/tests/test_ai_agent_interaction_learning_proof.py

@@ -13,9 +13,9 @@ def test_load_latest_ai_agent_interaction_learning_proof_reads_committed_snapsho
     data = load_latest_ai_agent_interaction_learning_proof()
 
     assert data["schema_version"] == "ai_agent_interaction_learning_proof_v1"
-    assert data["program_status"]["overall_completion_percent"] == 88
-    assert data["program_status"]["current_task_id"] == "P2-403F"
-    assert data["program_status"]["next_task_id"] == "P2-403G"
+    assert data["program_status"]["overall_completion_percent"] == 94
+    assert data["program_status"]["current_task_id"] == "P2-403G"
+    assert data["program_status"]["next_task_id"] == "P2-403H"
     assert data["program_status"]["read_only_mode"] is True
     assert data["program_status"]["runtime_authority"] == "proof_surface_only_no_live_worker"
     assert data["live_truth"]["runtime_loop_enabled"] is False

apps/api/tests/test_ai_agent_interaction_learning_proof_api.py

@@ -16,9 +16,9 @@ def test_ai_agent_interaction_learning_proof_endpoint_returns_committed_snapshot
     assert response.status_code == 200
     data = response.json()
     assert data["schema_version"] == "ai_agent_interaction_learning_proof_v1"
-    assert data["program_status"]["overall_completion_percent"] == 88
-    assert data["program_status"]["current_task_id"] == "P2-403F"
-    assert data["program_status"]["next_task_id"] == "P2-403G"
+    assert data["program_status"]["overall_completion_percent"] == 94
+    assert data["program_status"]["current_task_id"] == "P2-403G"
+    assert data["program_status"]["next_task_id"] == "P2-403H"
     assert data["program_status"]["read_only_mode"] is True
     assert data["live_truth"]["runtime_loop_enabled"] is False
     assert data["live_truth"]["active_live_agent_sessions"] == 0

apps/api/tests/test_ai_agent_proactive_operations_contract.py

@@ -14,8 +14,8 @@ def test_load_latest_ai_agent_proactive_operations_contract_reads_committed_snap
 
     assert data["schema_version"] == "ai_agent_proactive_operations_contract_v1"
     assert data["program_status"]["overall_completion_percent"] == 100
-    assert data["program_status"]["current_task_id"] == "P2-403F"
-    assert data["program_status"]["next_task_id"] == "P2-403G"
+    assert data["program_status"]["current_task_id"] == "P2-403G"
+    assert data["program_status"]["next_task_id"] == "P2-403H"
     assert data["program_status"]["read_only_mode"] is True
     assert data["program_status"]["runtime_authority"] == "contract_only_no_version_or_runtime_update"
     assert data["approval_boundaries"]["runtime_version_update_allowed"] is False
@@ -25,7 +25,7 @@ def test_load_latest_ai_agent_proactive_operations_contract_reads_committed_snap
     assert data["approval_boundaries"]["telegram_direct_send_allowed"] is False
     assert data["rollups"]["version_domain_count"] == len(data["version_lifecycle_domains"]) == 12
     assert data["rollups"]["delegable_capability_count"] == len(data["delegable_capabilities"]) == 24
-    assert data["rollups"]["rollout_task_count"] == len(data["rollout_tasks"]) == 13
+    assert data["rollups"]["rollout_task_count"] == len(data["rollout_tasks"]) == 14
     assert data["rollups"]["auto_execute_allowed_count"] == 0
     assert any(domain["domain_id"] == "ai_agents_models" for domain in data["version_lifecycle_domains"])
     assert any(

apps/api/tests/test_ai_agent_proactive_operations_contract_api.py

@@ -17,8 +17,8 @@ def test_ai_agent_proactive_operations_contract_endpoint_returns_committed_snaps
     data = response.json()
     assert data["schema_version"] == "ai_agent_proactive_operations_contract_v1"
     assert data["program_status"]["overall_completion_percent"] == 100
-    assert data["program_status"]["current_task_id"] == "P2-403F"
-    assert data["program_status"]["next_task_id"] == "P2-403G"
+    assert data["program_status"]["current_task_id"] == "P2-403G"
+    assert data["program_status"]["next_task_id"] == "P2-403H"
     assert data["program_status"]["read_only_mode"] is True
     assert data["approval_boundaries"]["runtime_version_update_allowed"] is False
     assert data["approval_boundaries"]["package_upgrade_allowed"] is False
@@ -26,7 +26,7 @@ def test_ai_agent_proactive_operations_contract_endpoint_returns_committed_snaps
     assert data["approval_boundaries"]["telegram_direct_send_allowed"] is False
     assert data["rollups"]["version_domain_count"] == 12
     assert data["rollups"]["delegable_capability_count"] == 24
-    assert data["rollups"]["rollout_task_count"] == 13
+    assert data["rollups"]["rollout_task_count"] == 14
     assert data["rollups"]["auto_execute_allowed_count"] == 0
     assert any(domain["domain_id"] == "host_os_packages" for domain in data["version_lifecycle_domains"])
     assert any(

apps/api/tests/test_ai_agent_runtime_write_gate_review.py

@@ -0,0 +1,74 @@
+import copy
+import json
+
+import pytest
+
+from src.services.ai_agent_runtime_write_gate_review import (
+    load_latest_ai_agent_runtime_write_gate_review,
+)
+
+
+def _write_snapshot(tmp_path, payload):
+    path = tmp_path / "ai_agent_runtime_write_gate_review_2026-06-12.json"
+    path.write_text(json.dumps(payload), encoding="utf-8")
+    return path
+
+
+def test_load_latest_ai_agent_runtime_write_gate_review():
+    data = load_latest_ai_agent_runtime_write_gate_review()
+
+    assert data["schema_version"] == "ai_agent_runtime_write_gate_review_v1"
+    assert data["program_status"]["current_task_id"] == "P2-403G"
+    assert data["program_status"]["next_task_id"] == "P2-403H"
+    assert data["program_status"]["overall_completion_percent"] == 94
+    assert data["runtime_write_truth"]["runtime_write_allowed"] is False
+    assert data["runtime_write_truth"]["dual_approval_required"] is True
+    assert data["runtime_write_truth"]["dual_approval_received_count"] == 0
+    assert data["runtime_write_truth"]["dry_run_hash_verified_count"] == 0
+    assert data["runtime_write_truth"]["post_write_verifier_pass_count"] == 0
+    assert data["rollups"]["write_target_count"] == len(data["write_targets"])
+    assert data["rollups"]["approval_gate_count"] == len(data["approval_gates"])
+    assert data["rollups"]["live_write_count_total"] == 0
+
+
+def test_rejects_runtime_write_enabled(tmp_path):
+    data = load_latest_ai_agent_runtime_write_gate_review()
+    bad = copy.deepcopy(data)
+    bad["runtime_write_truth"]["runtime_write_allowed"] = True
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="runtime write flags"):
+        load_latest_ai_agent_runtime_write_gate_review(tmp_path)
+
+
+def test_rejects_dual_approval_count_increment(tmp_path):
+    data = load_latest_ai_agent_runtime_write_gate_review()
+    bad = copy.deepcopy(data)
+    bad["runtime_write_truth"]["dual_approval_received_count"] = 1
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="write gate counts"):
+        load_latest_ai_agent_runtime_write_gate_review(tmp_path)
+
+
+def test_rejects_missing_required_review_field(tmp_path):
+    data = load_latest_ai_agent_runtime_write_gate_review()
+    bad = copy.deepcopy(data)
+    bad["write_gate_review"]["required_fields"] = [
+        field for field in bad["write_gate_review"]["required_fields"] if field != "dry_run_preview_hash"
+    ]
+    bad["rollups"]["required_field_count"] = len(bad["write_gate_review"]["required_fields"])
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="missing required fields"):
+        load_latest_ai_agent_runtime_write_gate_review(tmp_path)
+
+
+def test_rejects_rollup_mismatch(tmp_path):
+    data = load_latest_ai_agent_runtime_write_gate_review()
+    bad = copy.deepcopy(data)
+    bad["rollups"]["approval_gate_count"] = 999
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="rollup counts"):
+        load_latest_ai_agent_runtime_write_gate_review(tmp_path)

apps/api/tests/test_ai_agent_runtime_write_gate_review_api.py

@@ -0,0 +1,20 @@
+from fastapi.testclient import TestClient
+
+from src.main import app
+
+
+def test_get_ai_agent_runtime_write_gate_review_api():
+    client = TestClient(app)
+    response = client.get("/api/v1/agents/agent-runtime-write-gate-review")
+
+    assert response.status_code == 200
+    data = response.json()
+    assert data["schema_version"] == "ai_agent_runtime_write_gate_review_v1"
+    assert data["program_status"]["current_task_id"] == "P2-403G"
+    assert data["program_status"]["next_task_id"] == "P2-403H"
+    assert data["program_status"]["overall_completion_percent"] == 94
+    assert data["runtime_write_truth"]["runtime_write_allowed"] is False
+    assert data["runtime_write_truth"]["dual_approval_received_count"] == 0
+    assert data["runtime_write_truth"]["dry_run_hash_verified_count"] == 0
+    assert data["runtime_write_truth"]["post_write_verifier_pass_count"] == 0
+    assert data["rollups"]["live_write_count_total"] == 0