feat(governance): 新增 runtime worker shadow gate

apps/api/src/services/ai_agent_runtime_worker_shadow_gate.py

@@ -0,0 +1,266 @@
+"""
+AI Agent runtime worker shadow gate snapshot.
+
+Loads the latest committed P2-404 shadow/no-write execution evidence gate.
+This module validates repo-committed evidence only; it never starts workers,
+writes Telegram Gateway queues, sends Telegram messages, reads secrets, or
+writes production targets.
+"""
+
+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_worker_shadow_gate_*.json"
+_SCHEMA_VERSION = "ai_agent_runtime_worker_shadow_gate_v1"
+_RUNTIME_AUTHORITY = "runtime_worker_shadow_no_write_execution_evidence_gate_only_no_live_send_or_write"
+
+
+def load_latest_ai_agent_runtime_worker_shadow_gate(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed AI Agent runtime worker shadow gate."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no AI Agent runtime worker shadow 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")
+    _require_schema(payload, str(latest))
+    _require_no_write_boundaries(payload, str(latest))
+    _require_shadow_candidates(payload, str(latest))
+    _require_replay_contract(payload, str(latest))
+    _require_verifier_contract(payload, str(latest))
+    _require_agent_roles(payload, str(latest))
+    _require_redaction_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") != _RUNTIME_AUTHORITY:
+        raise ValueError(f"{label}: runtime_authority must remain {_RUNTIME_AUTHORITY}")
+    if status.get("current_task_id") != "P2-404":
+        raise ValueError(f"{label}: current_task_id must be P2-404")
+    if status.get("next_task_id") != "P2-101":
+        raise ValueError(f"{label}: next_task_id must be P2-101")
+
+
+def _require_no_write_boundaries(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("shadow_gate_truth") or {}
+    required_true = {
+        "shadow_worker_evidence_gate_ready",
+        "promotion_from_fixture_readback_ready",
+        "no_write_replay_plan_ready",
+        "action_candidate_selection_ready",
+        "mcp_evidence_reuse_ready",
+        "verifier_dry_run_binding_ready",
+        "failure_lane_ready",
+        "operator_review_packet_ready",
+    }
+    missing = sorted(field for field in required_true if truth.get(field) is not True)
+    if missing:
+        raise ValueError(f"{label}: shadow readiness flags must remain true: {missing}")
+
+    required_false = {
+        "production_delivery_enabled",
+        "gateway_queue_write_enabled",
+        "telegram_send_enabled",
+        "telegram_bot_api_call_enabled",
+        "delivery_receipt_write_enabled",
+        "shadow_worker_live_enabled",
+        "ai_runtime_worker_enabled",
+        "medium_low_auto_worker_enabled",
+        "post_action_verifier_live_readback_enabled",
+        "production_write_enabled",
+        "secret_value_read_enabled",
+        "work_window_transcript_display_allowed",
+    }
+    unsafe = sorted(field for field in required_false if truth.get(field) is not False)
+    if unsafe:
+        raise ValueError(f"{label}: live write/send/runtime flags must remain false: {unsafe}")
+
+    zero_counts = {
+        "shadow_worker_live_run_count_24h",
+        "gateway_queue_write_count_24h",
+        "telegram_send_count_24h",
+        "telegram_bot_api_call_count_24h",
+        "delivery_receipt_write_count_24h",
+        "ai_runtime_worker_run_count_24h",
+        "medium_low_auto_execution_count_24h",
+        "post_action_verifier_live_readback_count_24h",
+        "production_write_count_24h",
+    }
+    non_zero = sorted(field for field in zero_counts if truth.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: live write/send/runtime counts must remain zero: {non_zero}")
+
+
+def _require_shadow_candidates(payload: dict[str, Any], label: str) -> None:
+    candidates = payload.get("shadow_worker_candidates") or []
+    candidate_ids = {candidate.get("candidate_id") for candidate in candidates}
+    required = {
+        "shadow_report_daily_digest_candidate",
+        "shadow_report_weekly_digest_candidate",
+        "shadow_report_monthly_digest_candidate",
+        "shadow_medium_low_noop_repair_candidate",
+        "shadow_post_action_verifier_candidate",
+    }
+    if candidate_ids != required:
+        raise ValueError(f"{label}: shadow candidates must match {sorted(required)}")
+
+    for candidate in candidates:
+        candidate_id = candidate.get("candidate_id")
+        if not _is_redacted_sha256(candidate.get("promotion_hash")):
+            raise ValueError(f"{label}: candidate {candidate_id} must expose a redacted sha256 promotion_hash")
+        if candidate.get("writes_production") is not False:
+            raise ValueError(f"{label}: candidate {candidate_id} must not write production")
+        if candidate.get("sends_telegram") is not False:
+            raise ValueError(f"{label}: candidate {candidate_id} must not send Telegram")
+        if candidate.get("reads_secret_value") is not False:
+            raise ValueError(f"{label}: candidate {candidate_id} must not read secret value")
+        if candidate.get("live_shadow_run_count_24h") != 0:
+            raise ValueError(f"{label}: candidate {candidate_id} live_shadow_run_count_24h must remain zero")
+
+
+def _require_replay_contract(payload: dict[str, Any], label: str) -> None:
+    replays = payload.get("no_write_execution_replays") or []
+    replay_ids = {replay.get("replay_id") for replay in replays}
+    required = {
+        "daily_report_shadow_replay",
+        "weekly_report_shadow_replay",
+        "medium_low_noop_shadow_replay",
+        "verifier_shadow_replay",
+    }
+    if replay_ids != required:
+        raise ValueError(f"{label}: no-write replays must match {sorted(required)}")
+
+    for replay in replays:
+        replay_id = replay.get("replay_id")
+        if not _is_redacted_sha256(replay.get("evidence_hash")):
+            raise ValueError(f"{label}: replay {replay_id} must expose a redacted sha256 evidence_hash")
+        if replay.get("verifier_bound") is not True:
+            raise ValueError(f"{label}: replay {replay_id} must remain verifier bound")
+        if replay.get("writes_result") is not False:
+            raise ValueError(f"{label}: replay {replay_id} must not write result")
+        if replay.get("production_side_effect_count") != 0:
+            raise ValueError(f"{label}: replay {replay_id} production_side_effect_count must remain zero")
+
+
+def _require_verifier_contract(payload: dict[str, Any], label: str) -> None:
+    cases = payload.get("verifier_shadow_cases") or []
+    case_ids = {case.get("case_id") for case in cases}
+    required = {
+        "shadow_promotion_verifier_case",
+        "queue_write_block_verifier_case",
+        "telegram_send_block_verifier_case",
+        "production_write_block_verifier_case",
+    }
+    if case_ids != required:
+        raise ValueError(f"{label}: verifier shadow cases must match {sorted(required)}")
+
+    for case in cases:
+        case_id = case.get("case_id")
+        if not _is_redacted_sha256(case.get("evidence_hash")):
+            raise ValueError(f"{label}: verifier case {case_id} must expose a redacted sha256 evidence_hash")
+        if case.get("live_readback_enabled") is not False:
+            raise ValueError(f"{label}: verifier case {case_id} must not run live readback")
+        if case.get("writes_result") is not False:
+            raise ValueError(f"{label}: verifier case {case_id} must not write result")
+        if case.get("requires_secret_value") is not False:
+            raise ValueError(f"{label}: verifier case {case_id} must not require secret value")
+
+
+def _require_agent_roles(payload: dict[str, Any], label: str) -> None:
+    roles = payload.get("agent_shadow_roles") or []
+    agents = {role.get("agent_id") for role in roles}
+    if agents != {"openclaw", "hermes", "nemotron"}:
+        raise ValueError(f"{label}: shadow roles must include OpenClaw, Hermes, and NemoTron")
+    for role in roles:
+        if role.get("live_action_count_24h") != 0:
+            raise ValueError(f"{label}: agent {role.get('agent_id')} live_action_count_24h must remain zero")
+
+
+def _require_redaction_contract(payload: dict[str, Any], label: str) -> None:
+    contract = payload.get("display_redaction_contract") or {}
+    required_false = {
+        "raw_report_payload_display_allowed",
+        "raw_telegram_payload_display_allowed",
+        "raw_shadow_payload_display_allowed",
+        "private_reasoning_display_allowed",
+        "secret_value_display_allowed",
+        "work_window_transcript_display_allowed",
+    }
+    if contract.get("redaction_required") is not True:
+        raise ValueError(f"{label}: display redaction must remain required")
+    unsafe = sorted(field for field in required_false if contract.get(field) is not False)
+    if unsafe:
+        raise ValueError(f"{label}: display redaction fields must remain false: {unsafe}")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    rollups = payload.get("rollups") or {}
+    truth = payload.get("shadow_gate_truth") or {}
+    candidates = payload.get("shadow_worker_candidates") or []
+    replays = payload.get("no_write_execution_replays") or []
+    cases = payload.get("verifier_shadow_cases") or []
+    roles = payload.get("agent_shadow_roles") or []
+    checkpoints = payload.get("operator_checkpoints") or []
+
+    expected = {
+        "shadow_candidate_count": len(candidates),
+        "passed_no_write_candidate_count": sum(1 for item in candidates if item.get("shadow_status") == "passed_no_write_replay"),
+        "blocked_candidate_count": sum(1 for item in candidates if item.get("shadow_status") == "blocked_by_runtime_gate"),
+        "needs_owner_review_candidate_count": sum(1 for item in candidates if item.get("shadow_status") == "needs_owner_review"),
+        "no_write_replay_count": len(replays),
+        "passed_no_write_replay_count": sum(1 for item in replays if item.get("replay_status") == "passed_no_write"),
+        "verifier_shadow_case_count": len(cases),
+        "agent_role_count": len(roles),
+        "operator_checkpoint_count": len(checkpoints),
+    }
+    mismatches = sorted(field for field, value in expected.items() if rollups.get(field) != value)
+    if mismatches:
+        raise ValueError(f"{label}: rollup counts must match source arrays: {mismatches}")
+
+    approval_ids = sorted(item.get("checkpoint_id") for item in checkpoints if item.get("approval_required") is True)
+    if sorted(rollups.get("approval_required_checkpoint_ids") or []) != approval_ids:
+        raise ValueError(f"{label}: approval_required_checkpoint_ids must match operator checkpoints")
+
+    zero_pairs = {
+        "shadow_worker_live_run_count": truth.get("shadow_worker_live_run_count_24h"),
+        "gateway_queue_write_count": truth.get("gateway_queue_write_count_24h"),
+        "telegram_send_count": truth.get("telegram_send_count_24h"),
+        "telegram_bot_api_call_count": truth.get("telegram_bot_api_call_count_24h"),
+        "delivery_receipt_write_count": truth.get("delivery_receipt_write_count_24h"),
+        "ai_runtime_worker_run_count": truth.get("ai_runtime_worker_run_count_24h"),
+        "medium_low_auto_execution_count": truth.get("medium_low_auto_execution_count_24h"),
+        "post_action_verifier_live_readback_count": truth.get("post_action_verifier_live_readback_count_24h"),
+        "production_write_count": truth.get("production_write_count_24h"),
+    }
+    non_zero = sorted(field for field, value in zero_pairs.items() if rollups.get(field) != 0 or value != 0)
+    if non_zero:
+        raise ValueError(f"{label}: rollup live counts must remain zero: {non_zero}")
+
+
+def _is_redacted_sha256(value: Any) -> bool:
+    if not isinstance(value, str):
+        return False
+    if not value.startswith("sha256:"):
+        return False
+    digest = value.removeprefix("sha256:")
+    return len(digest) == 64 and all(char in "0123456789abcdef" for char in digest)