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

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")