feat(governance): 新增報表 fixture readback 證據包

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

@@ -91,6 +91,9 @@ from src.services.ai_agent_report_automation_review import (
 from src.services.ai_agent_report_runtime_dry_run import (
     load_latest_ai_agent_report_runtime_dry_run,
 )
+from src.services.ai_agent_report_runtime_fixture_readback import (
+    load_latest_ai_agent_report_runtime_fixture_readback,
+)
 from src.services.ai_agent_report_runtime_readiness import (
     load_latest_ai_agent_report_runtime_readiness,
 )
@@ -973,6 +976,34 @@ async def get_agent_report_runtime_dry_run() -> dict[str, Any]:
         ) from exc
 
 
+@router.get(
+    "/agent-report-runtime-fixture-readback",
+    response_model=dict[str, Any],
+    summary="取得 AI Agent 報表 runtime fixture readback 證據包",
+    description=(
+        "讀取最新已提交的 P2-403N fixture smoke、Telegram Gateway queue preview readback "
+        "與 verifier dry-run 證據包；此端點不排程實發、不送 Telegram、不呼叫 Bot API、"
+        "不寫 Gateway queue、不寫讀報回執、不啟動 AI runtime worker、不啟動中低風險 auto worker、"
+        "不執行 verifier live readback、不寫 production target、不讀 secret、不回傳內部對話內容。"
+    ),
+)
+async def get_agent_report_runtime_fixture_readback() -> dict[str, Any]:
+    """Return the latest read-only AI Agent fixture readback package."""
+    try:
+        return await asyncio.to_thread(load_latest_ai_agent_report_runtime_fixture_readback)
+    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_runtime_fixture_readback_invalid", error=str(exc))
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="AI Agent 報表 runtime fixture readback 證據包無效",
+        ) from exc
+
+
 @router.get(
     "/agent-owner-approved-fixture-dry-run",
     response_model=dict[str, Any],

apps/api/src/services/ai_agent_report_runtime_fixture_readback.py

@@ -0,0 +1,269 @@
+"""
+AI Agent report runtime fixture readback snapshot.
+
+Loads the latest committed P2-403N fixture smoke / queue preview readback /
+verifier dry-run contract. This module only validates repo-committed evidence
+and never writes Telegram Gateway queues, sends Telegram messages, starts AI
+workers, runs live verifiers, or reads secrets.
+"""
+
+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_runtime_fixture_readback_*.json"
+_SCHEMA_VERSION = "ai_agent_report_runtime_fixture_readback_v1"
+_RUNTIME_AUTHORITY = "fixture_smoke_queue_preview_readback_verifier_dry_run_only_no_live_send_or_write"
+
+
+def load_latest_ai_agent_report_runtime_fixture_readback(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed AI Agent fixture readback snapshot."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no AI Agent report runtime fixture readback 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_fixture_contract(payload, str(latest))
+    _require_queue_readback_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") != "P2-403N":
+        raise ValueError(f"{label}: current_task_id must be P2-403N")
+    if status.get("next_task_id") != "P2-404":
+        raise ValueError(f"{label}: next_task_id must be P2-404")
+
+
+def _require_no_write_boundaries(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("fixture_readback_truth") or {}
+    required_true = {
+        "fixture_smoke_package_ready",
+        "report_snapshot_hash_ready",
+        "telegram_queue_preview_readback_ready",
+        "readback_verifier_dry_run_ready",
+        "redaction_assertions_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}: fixture readiness flags must remain true: {missing}")
+
+    required_false = {
+        "production_delivery_enabled",
+        "telegram_gateway_queue_write_enabled",
+        "telegram_send_enabled",
+        "telegram_bot_api_call_enabled",
+        "delivery_receipt_write_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 = {
+        "live_report_delivery_count_24h",
+        "telegram_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_fixture_contract(payload: dict[str, Any], label: str) -> None:
+    fixtures = payload.get("fixture_smoke_results") or []
+    fixture_ids = {fixture.get("fixture_id") for fixture in fixtures}
+    required = {
+        "report_run_snapshot_fixture",
+        "telegram_digest_payload_fixture",
+        "queue_preview_redaction_fixture",
+        "receipt_redaction_fixture",
+        "medium_low_noop_fixture",
+    }
+    if fixture_ids != required:
+        raise ValueError(f"{label}: fixture smoke results must match {sorted(required)}")
+
+    for fixture in fixtures:
+        fixture_id = fixture.get("fixture_id")
+        if not _is_redacted_sha256(fixture.get("output_hash")):
+            raise ValueError(f"{label}: fixture {fixture_id} must expose a redacted sha256 output_hash")
+        if fixture.get("writes_production") is not False:
+            raise ValueError(f"{label}: fixture {fixture_id} must not write production")
+        if fixture.get("sends_telegram") is not False:
+            raise ValueError(f"{label}: fixture {fixture_id} must not send Telegram")
+        if fixture.get("reads_secret_value") is not False:
+            raise ValueError(f"{label}: fixture {fixture_id} must not read secret value")
+        if fixture.get("live_execution_count_24h") != 0:
+            raise ValueError(f"{label}: fixture {fixture_id} live_execution_count_24h must remain zero")
+
+
+def _require_queue_readback_contract(payload: dict[str, Any], label: str) -> None:
+    readbacks = payload.get("queue_preview_readbacks") or []
+    readback_ids = {readback.get("readback_id") for readback in readbacks}
+    if readback_ids != {
+        "daily_report_digest_readback",
+        "weekly_report_digest_readback",
+        "monthly_report_digest_readback",
+    }:
+        raise ValueError(f"{label}: queue preview readbacks must cover daily, weekly, monthly")
+
+    for readback in readbacks:
+        readback_id = readback.get("readback_id")
+        if readback.get("recipient_room") != "AwoooI SRE 戰情室":
+            raise ValueError(f"{label}: readback {readback_id} must target AwoooI SRE 戰情室")
+        if readback.get("secret_ref") != "SRE_GROUP_CHAT_ID":
+            raise ValueError(f"{label}: readback {readback_id} must only reference SRE_GROUP_CHAT_ID")
+        if not _is_redacted_sha256(readback.get("preview_hash")):
+            raise ValueError(f"{label}: readback {readback_id} must expose a redacted sha256 preview_hash")
+        if readback.get("payload_redacted") is not True:
+            raise ValueError(f"{label}: readback {readback_id} payload must remain redacted")
+        if readback.get("gateway_queue_write_enabled") is not False:
+            raise ValueError(f"{label}: readback {readback_id} must not write Gateway queue")
+        if readback.get("telegram_send_enabled") is not False:
+            raise ValueError(f"{label}: readback {readback_id} must not send Telegram")
+        if readback.get("direct_bot_api_allowed") is not False:
+            raise ValueError(f"{label}: readback {readback_id} must not allow direct Bot API")
+        if readback.get("queue_write_count_24h") != 0:
+            raise ValueError(f"{label}: readback {readback_id} queue_write_count_24h must remain zero")
+
+
+def _require_verifier_contract(payload: dict[str, Any], label: str) -> None:
+    cases = payload.get("verifier_dry_run_cases") or []
+    case_ids = {case.get("case_id") for case in cases}
+    required = {
+        "report_snapshot_verifier_dry_run",
+        "gateway_preview_verifier_dry_run",
+        "receipt_redaction_verifier_dry_run",
+        "medium_low_noop_verifier_dry_run",
+    }
+    if case_ids != required:
+        raise ValueError(f"{label}: verifier dry-run 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_fixture_roles") or []
+    agents = {role.get("agent_id") for role in roles}
+    if agents != {"openclaw", "hermes", "nemotron"}:
+        raise ValueError(f"{label}: fixture 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",
+        "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("fixture_readback_truth") or {}
+    fixtures = payload.get("fixture_smoke_results") or []
+    readbacks = payload.get("queue_preview_readbacks") or []
+    cases = payload.get("verifier_dry_run_cases") or []
+    roles = payload.get("agent_fixture_roles") or []
+    checkpoints = payload.get("operator_checkpoints") or []
+
+    expected = {
+        "fixture_smoke_count": len(fixtures),
+        "passed_fixture_smoke_count": sum(
+            1
+            for fixture in fixtures
+            if fixture.get("smoke_status") in {"passed_no_write", "passed_redaction"}
+        ),
+        "queue_preview_readback_count": len(readbacks),
+        "verifier_dry_run_case_count": len(cases),
+        "agent_role_count": len(roles),
+        "operator_checkpoint_count": len(checkpoints),
+        "live_report_delivery_count": truth.get("live_report_delivery_count_24h"),
+        "telegram_gateway_queue_write_count": truth.get("telegram_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"),
+    }
+    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(
+        checkpoint.get("checkpoint_id")
+        for checkpoint in checkpoints
+        if checkpoint.get("approval_required") is True
+    )
+    if sorted(rollups.get("approval_required_checkpoint_ids") or []) != approval_required:
+        raise ValueError(f"{label}: approval_required_checkpoint_ids mismatch")
+
+
+def _is_redacted_sha256(value: Any) -> bool:
+    if not isinstance(value, str) or not value.startswith("sha256:"):
+        return False
+    digest = value.removeprefix("sha256:")
+    return len(digest) == 64 and all(char in "0123456789abcdef" for char in digest)