feat(governance): 新增 Agent 日週月報風險自動化審查

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

@@ -85,6 +85,9 @@ 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_report_automation_review import (
+    load_latest_ai_agent_report_automation_review,
+)
 from src.services.ai_agent_report_truth_actionability_review import (
     load_latest_ai_agent_report_truth_actionability_review,
 )
@@ -881,6 +884,33 @@ async def get_agent_report_truth_actionability_review() -> dict[str, Any]:
         ) from exc
 
 
+@router.get(
+    "/agent-report-automation-review",
+    response_model=dict[str, Any],
+    summary="取得 AI Agent 日週月報與風險自動化 review",
+    description=(
+        "讀取最新已提交的 AI Agent 日報、週報、月報、Agent 工作量、圖表化報告、"
+        "AI 分析建議與高/中/低風險自動化政策；此端點不排程實發、不送 Telegram、"
+        "不啟動中低風險自動執行器、不執行生產優化、不讀 secret、不回傳內部工作視窗對話。"
+    ),
+)
+async def get_agent_report_automation_review() -> dict[str, Any]:
+    """Return the latest read-only AI Agent report automation review."""
+    try:
+        return await asyncio.to_thread(load_latest_ai_agent_report_automation_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_report_automation_review_invalid", error=str(exc))
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="AI Agent 日週月報與風險自動化 review 無效",
+        ) from exc
+
+
 @router.get(
     "/agent-owner-approved-fixture-dry-run",
     response_model=dict[str, Any],

apps/api/src/services/ai_agent_report_automation_review.py

@@ -0,0 +1,154 @@
+"""
+AI Agent report automation review snapshot.
+
+Loads the latest committed P2-403J daily / weekly / monthly report, workload,
+chart, and risk-tier automation policy review. This module never schedules a
+live report, sends Telegram, writes optimization changes, or starts an
+automation worker.
+"""
+
+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_report_automation_review_*.json"
+_SCHEMA_VERSION = "ai_agent_report_automation_review_v1"
+
+
+def load_latest_ai_agent_report_automation_review(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed AI Agent report automation review snapshot."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no AI Agent report automation 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_report_contract(payload, str(latest))
+    _require_runtime_boundaries(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") != "reporting_and_risk_policy_review_only_no_live_execution":
+        raise ValueError(
+            f"{label}: runtime_authority must remain reporting_and_risk_policy_review_only_no_live_execution"
+        )
+
+
+def _require_report_contract(payload: dict[str, Any], label: str) -> None:
+    cadences = payload.get("report_cadences") or []
+    cadence_ids = {cadence.get("cadence_id") for cadence in cadences}
+    if cadence_ids != {"daily", "weekly", "monthly"}:
+        raise ValueError(f"{label}: report cadences must include daily, weekly, monthly")
+
+    agent_ids = {agent.get("agent_id") for agent in payload.get("agent_workload_metrics") or []}
+    if agent_ids != {"openclaw", "hermes", "nemotron"}:
+        raise ValueError(f"{label}: workload metrics must include OpenClaw, Hermes, NemoTron")
+
+    if not payload.get("report_charts"):
+        raise ValueError(f"{label}: report charts must not be empty")
+    if not payload.get("analysis_recommendations"):
+        raise ValueError(f"{label}: analysis recommendations must not be empty")
+
+    risk_ids = {tier.get("risk_id") for tier in (payload.get("risk_tier_policy") or {}).get("risk_tiers") or []}
+    if not {"low", "medium", "high", "critical"}.issubset(risk_ids):
+        raise ValueError(f"{label}: risk tier policy must include low, medium, high, critical")
+
+
+def _require_runtime_boundaries(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("report_truth") or {}
+    if truth.get("high_risk_requires_approval") is not True:
+        raise ValueError(f"{label}: high risk approval gate must remain true")
+    if truth.get("medium_low_auto_policy_defined") is not True:
+        raise ValueError(f"{label}: medium / low auto policy must be defined")
+
+    zero_counts = {
+        "report_delivery_count_24h",
+        "report_read_receipt_count_24h",
+        "live_auto_optimization_count_24h",
+        "live_medium_low_auto_execution_count_24h",
+    }
+    non_zero = sorted(key for key in zero_counts if truth.get(key) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: live report / automation counts must remain zero: {non_zero}")
+
+    false_flags = {
+        "report_delivery_enabled",
+        "ai_analysis_after_report_enabled",
+        "medium_low_auto_execution_enabled",
+    }
+    unsafe_truth = sorted(flag for flag in false_flags if truth.get(flag) is not False)
+    if unsafe_truth:
+        raise ValueError(f"{label}: live report automation flags must remain false: {unsafe_truth}")
+
+    boundaries = payload.get("approval_boundaries") or {}
+    unsafe_boundaries = sorted(
+        key
+        for key, value in boundaries.items()
+        if key != "high_risk_requires_human_approval" and value is not False
+    )
+    if unsafe_boundaries:
+        raise ValueError(f"{label}: approval boundaries must remain false: {unsafe_boundaries}")
+    if boundaries.get("high_risk_requires_human_approval") is not True:
+        raise ValueError(f"{label}: high_risk_requires_human_approval must remain true")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    rollups = payload.get("rollups") or {}
+    workload = payload.get("agent_workload_metrics") or []
+    recommendations = payload.get("analysis_recommendations") or []
+    charts = payload.get("report_charts") or []
+    cadences = payload.get("report_cadences") or []
+
+    expected = {
+        "report_cadence_count": len(cadences),
+        "agent_count": len(workload),
+        "chart_count": len(charts),
+        "recommendation_count": len(recommendations),
+        "workload_unit_total": sum(item.get("work_units_total", 0) for item in workload),
+        "workload_done_total": sum(item.get("work_units_done", 0) for item in workload),
+        "workload_waiting_approval_total": sum(item.get("work_units_waiting_approval", 0) for item in workload),
+        "live_report_delivery_count": 0,
+        "live_auto_optimization_count": 0,
+    }
+    for risk in ("low", "medium", "high", "critical"):
+        expected[f"{risk}_risk_recommendation_count"] = len(
+            [item for item in recommendations if item.get("risk_tier") == risk]
+        )
+
+    mismatched = {
+        key: {"expected": value, "actual": rollups.get(key)}
+        for key, value in expected.items()
+        if rollups.get(key) != value
+    }
+    if mismatched:
+        raise ValueError(f"{label}: rollup counts must match payload sections: {mismatched}")
+
+    approval_required = sorted(
+        item.get("recommendation_id")
+        for item in recommendations
+        if item.get("approval_required") is True
+    )
+    if sorted(rollups.get("approval_required_recommendation_ids") or []) != approval_required:
+        raise ValueError(f"{label}: approval_required_recommendation_ids mismatch")
+    if rollups.get("current_auto_execution_enabled_count") != 0:
+        raise ValueError(f"{label}: current auto execution count must remain zero")

apps/api/tests/test_ai_agent_interaction_learning_proof.py

@@ -13,7 +13,7 @@ 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"] == 99
+    assert data["program_status"]["overall_completion_percent"] == 100
     assert data["program_status"]["current_task_id"] == "P2-403J"
     assert data["program_status"]["next_task_id"] == "P2-403K"
     assert data["program_status"]["read_only_mode"] is True
@@ -26,10 +26,10 @@ def test_load_latest_ai_agent_interaction_learning_proof_reads_committed_snapsho
     assert data["frontend_redaction"]["raw_prompt_display_allowed"] is False
     assert data["approval_boundaries"]["runtime_worker_allowed"] is False
     assert data["approval_boundaries"]["telegram_direct_send_allowed"] is False
-    assert data["rollups"]["proof_level_count"] == len(data["proof_ladder"]) == 8
-    assert data["rollups"]["signal_count"] == len(data["proof_signals"]) == 10
+    assert data["rollups"]["proof_level_count"] == len(data["proof_ladder"]) == 9
+    assert data["rollups"]["signal_count"] == len(data["proof_signals"]) == 11
     assert data["rollups"]["operator_surface_count"] == len(data["operator_surfaces"]) == 5
-    assert data["rollups"]["runtime_gate_count"] == len(data["runtime_gates"]) == 6
+    assert data["rollups"]["runtime_gate_count"] == len(data["runtime_gates"]) == 7
     assert data["rollups"]["live_signal_count"] == 0
     assert data["rollups"]["live_pending_level_ids"] == []
     assert {lane["agent_id"] for lane in data["agent_lanes"]} == {

apps/api/tests/test_ai_agent_interaction_learning_proof_api.py

@@ -16,7 +16,7 @@ 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"] == 99
+    assert data["program_status"]["overall_completion_percent"] == 100
     assert data["program_status"]["current_task_id"] == "P2-403J"
     assert data["program_status"]["next_task_id"] == "P2-403K"
     assert data["program_status"]["read_only_mode"] is True
@@ -26,9 +26,9 @@ def test_ai_agent_interaction_learning_proof_endpoint_returns_committed_snapshot
     assert data["frontend_redaction"]["operator_conversation_display_allowed"] is False
     assert data["approval_boundaries"]["conversation_transcript_display_allowed"] is False
     assert data["approval_boundaries"]["telegram_direct_send_allowed"] is False
-    assert data["rollups"]["proof_level_count"] == 8
-    assert data["rollups"]["contract_ready_level_count"] == 6
-    assert data["rollups"]["signal_count"] == 10
+    assert data["rollups"]["proof_level_count"] == 9
+    assert data["rollups"]["contract_ready_level_count"] == 7
+    assert data["rollups"]["signal_count"] == 11
     assert data["rollups"]["live_signal_count"] == 0
     assert data["rollups"]["operator_surface_count"] == 5
     assert any(lane["agent_id"] == "openclaw" for lane in data["agent_lanes"])

apps/api/tests/test_ai_agent_report_automation_review.py

@@ -0,0 +1,91 @@
+import copy
+import json
+
+import pytest
+
+from src.services.ai_agent_report_automation_review import (
+    load_latest_ai_agent_report_automation_review,
+)
+
+
+def _write_snapshot(tmp_path, payload):
+    path = tmp_path / "ai_agent_report_automation_review_2026-06-12.json"
+    path.write_text(json.dumps(payload), encoding="utf-8")
+    return path
+
+
+def test_load_latest_ai_agent_report_automation_review():
+    data = load_latest_ai_agent_report_automation_review()
+
+    assert data["schema_version"] == "ai_agent_report_automation_review_v1"
+    assert data["program_status"]["current_task_id"] == "P2-403J"
+    assert data["program_status"]["next_task_id"] == "P2-403K"
+    assert data["program_status"]["overall_completion_percent"] == 100
+    assert data["report_truth"]["daily_report_ready"] is True
+    assert data["report_truth"]["weekly_report_ready"] is True
+    assert data["report_truth"]["monthly_report_ready"] is True
+    assert data["report_truth"]["medium_low_auto_policy_defined"] is True
+    assert data["report_truth"]["medium_low_auto_execution_enabled"] is False
+    assert data["report_truth"]["high_risk_requires_approval"] is True
+    assert data["report_truth"]["live_auto_optimization_count_24h"] == 0
+    assert data["rollups"]["report_cadence_count"] == 3
+    assert data["rollups"]["agent_count"] == 3
+    assert data["rollups"]["chart_count"] == 4
+    assert data["rollups"]["recommendation_count"] == 5
+    assert data["rollups"]["workload_unit_total"] == 91
+    assert data["rollups"]["current_auto_execution_enabled_count"] == 0
+    assert data["rollups"]["live_report_delivery_count"] == 0
+    assert data["rollups"]["live_auto_optimization_count"] == 0
+
+
+def test_rejects_missing_monthly_report(tmp_path):
+    data = load_latest_ai_agent_report_automation_review()
+    bad = copy.deepcopy(data)
+    bad["report_cadences"] = [
+        cadence for cadence in bad["report_cadences"] if cadence["cadence_id"] != "monthly"
+    ]
+    bad["rollups"]["report_cadence_count"] = len(bad["report_cadences"])
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="daily, weekly, monthly"):
+        load_latest_ai_agent_report_automation_review(tmp_path)
+
+
+def test_rejects_medium_low_auto_execution_enabled(tmp_path):
+    data = load_latest_ai_agent_report_automation_review()
+    bad = copy.deepcopy(data)
+    bad["report_truth"]["medium_low_auto_execution_enabled"] = True
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="live report automation flags"):
+        load_latest_ai_agent_report_automation_review(tmp_path)
+
+
+def test_rejects_high_risk_approval_disabled(tmp_path):
+    data = load_latest_ai_agent_report_automation_review()
+    bad = copy.deepcopy(data)
+    bad["report_truth"]["high_risk_requires_approval"] = False
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="high risk approval gate"):
+        load_latest_ai_agent_report_automation_review(tmp_path)
+
+
+def test_rejects_report_delivery_count(tmp_path):
+    data = load_latest_ai_agent_report_automation_review()
+    bad = copy.deepcopy(data)
+    bad["report_truth"]["report_delivery_count_24h"] = 1
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="live report / automation counts"):
+        load_latest_ai_agent_report_automation_review(tmp_path)
+
+
+def test_rejects_rollup_mismatch(tmp_path):
+    data = load_latest_ai_agent_report_automation_review()
+    bad = copy.deepcopy(data)
+    bad["rollups"]["workload_unit_total"] = 999
+    _write_snapshot(tmp_path, bad)
+
+    with pytest.raises(ValueError, match="rollup counts"):
+        load_latest_ai_agent_report_automation_review(tmp_path)

apps/api/tests/test_ai_agent_report_automation_review_api.py

@@ -0,0 +1,28 @@
+from fastapi.testclient import TestClient
+
+from src.main import app
+
+
+def test_get_ai_agent_report_automation_review_api():
+    client = TestClient(app)
+    response = client.get("/api/v1/agents/agent-report-automation-review")
+
+    assert response.status_code == 200
+    data = response.json()
+    assert data["schema_version"] == "ai_agent_report_automation_review_v1"
+    assert data["program_status"]["current_task_id"] == "P2-403J"
+    assert data["program_status"]["next_task_id"] == "P2-403K"
+    assert data["program_status"]["overall_completion_percent"] == 100
+    assert data["report_truth"]["daily_report_ready"] is True
+    assert data["report_truth"]["weekly_report_ready"] is True
+    assert data["report_truth"]["monthly_report_ready"] is True
+    assert data["report_truth"]["medium_low_auto_policy_defined"] is True
+    assert data["report_truth"]["medium_low_auto_execution_enabled"] is False
+    assert data["report_truth"]["high_risk_requires_approval"] is True
+    assert data["rollups"]["report_cadence_count"] == 3
+    assert data["rollups"]["agent_count"] == 3
+    assert data["rollups"]["chart_count"] == 4
+    assert data["rollups"]["recommendation_count"] == 5
+    assert data["rollups"]["workload_unit_total"] == 91
+    assert data["rollups"]["current_auto_execution_enabled_count"] == 0
+    assert data["rollups"]["live_auto_optimization_count"] == 0