feat(governance): 新增 runtime readback fixture approval

2026-06-13 17:07:14 +08:00
parent f8d6c0c388
commit 17815e5d20
14 changed files with 2066 additions and 11 deletions
--- a/apps/api/src/api/v1/agents.py
+++ b/apps/api/src/api/v1/agents.py
@@ -109,6 +109,9 @@ from src.services.ai_agent_runtime_readback_approval_package import (
 from src.services.ai_agent_runtime_readback_implementation_review import (
    load_latest_ai_agent_runtime_readback_implementation_review,
 )
+from src.services.ai_agent_runtime_readback_fixture_approval import (
+    load_latest_ai_agent_runtime_readback_fixture_approval,
+)
 from src.services.ai_agent_report_live_delivery_approval_package import (
    load_latest_ai_agent_report_live_delivery_approval_package,
 )
@@ -1392,6 +1395,36 @@ async def get_agent_report_live_delivery_approval_package() -> dict[str, Any]:
        ) from exc


+@router.get(
+    "/agent-runtime-readback-fixture-approval",
+    response_model=dict[str, Any],
+    summary="取得 AI Agent runtime readback fixture 批准包",
+    description=(
+        "讀取最新已提交的 P2-112 runtime readback fixture approval package；"
+        "此端點只回傳 fixture approval card、adapter contract、verifier fixture、"
+        "blocker mapping 與 operator action，不讀 canonical runtime target、不做 live query、"
+        "不執行 runtime readback、不寫 Gateway queue、不送 Telegram、不呼叫 Bot API、"
+        "不寫 report receipt、不寫 result capture、不寫 production target、不讀 secret。"
+    ),
+)
+async def get_agent_runtime_readback_fixture_approval() -> dict[str, Any]:
+    """Return the latest read-only runtime readback fixture approval package."""
+    try:
+        payload = await asyncio.to_thread(load_latest_ai_agent_runtime_readback_fixture_approval)
+        return redact_public_lan_topology(payload)
+    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_readback_fixture_approval_invalid", error=str(exc))
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="AI Agent runtime readback fixture approval 無效",
+        ) from exc
+
+
@router.get(
    "/agent-owner-approved-fixture-dry-run",
    response_model=dict[str, Any],
--- a/apps/api/src/services/ai_agent_runtime_readback_fixture_approval.py
+++ b/apps/api/src/services/ai_agent_runtime_readback_fixture_approval.py
@@ -0,0 +1,387 @@
+"""
+AI Agent runtime readback fixture approval snapshot.
+
+Loads the latest committed P2-112 fixture-only runtime readback approval package.
+This module validates committed evidence only; it never reads canonical runtime
+targets, performs live queries, executes runtime readback, writes Gateway queues,
+sends Telegram messages, calls Bot API, writes receipts/result captures, reads
+secrets, or performs destructive operations.
+"""
+
+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_readback_fixture_approval_*.json"
+_SCHEMA_VERSION = "ai_agent_runtime_readback_fixture_approval_v1"
+_RUNTIME_AUTHORITY = "runtime_readback_fixture_approval_only_no_canonical_target_or_live_query"
+
+
+def load_latest_ai_agent_runtime_readback_fixture_approval(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed runtime readback fixture approval package."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no AI Agent runtime readback fixture approval 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")
+
+    label = str(latest)
+    _require_schema(payload, label)
+    _require_prior(payload, label)
+    _require_truth(payload, label)
+    _require_cards(payload, label)
+    _require_contracts(payload, label)
+    _require_checks(payload, label)
+    _require_blockers(payload, label)
+    _require_actions(payload, label)
+    _require_display_redaction(payload, label)
+    _require_no_forbidden_display_terms(payload, label)
+    _require_rollup_consistency(payload, label)
+    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 {}
+    expected = {
+        "current_priority": "P2",
+        "current_task_id": "P2-112",
+        "next_task_id": "P2-113",
+        "read_only_mode": True,
+        "runtime_authority": _RUNTIME_AUTHORITY,
+        "overall_completion_percent": 100,
+    }
+    mismatches = _mismatches(status, expected)
+    if mismatches:
+        raise ValueError(f"{label}: program_status mismatch: {mismatches}")
+    if not status.get("status_note"):
+        raise ValueError(f"{label}: program_status.status_note is required")
+
+
+def _require_prior(payload: dict[str, Any], label: str) -> None:
+    runtime = payload.get("prior_runtime_review") or {}
+    runtime_expected = {
+        "implementation_review_schema_version": "ai_agent_runtime_readback_implementation_review_v1",
+        "implementation_review_card_count": 5,
+        "no_write_verifier_check_count": 5,
+        "implementation_blocker_count": 5,
+        "runtime_readback_execution_count": 0,
+        "live_query_count": 0,
+        "production_write_count": 0,
+    }
+    runtime_mismatches = _mismatches(runtime, runtime_expected)
+    if runtime_mismatches:
+        raise ValueError(f"{label}: prior_runtime_review mismatch: {runtime_mismatches}")
+    if not runtime.get("readiness_note"):
+        raise ValueError(f"{label}: prior_runtime_review.readiness_note is required")
+
+    delivery = payload.get("prior_delivery_approval") or {}
+    delivery_expected = {
+        "delivery_approval_schema_version": "ai_agent_report_live_delivery_approval_package_v1",
+        "delivery_approval_packet_count": 5,
+        "route_lock_gate_count": 4,
+        "payload_redaction_check_count": 5,
+        "dry_run_delivery_receipt_count": 4,
+        "telegram_send_count": 0,
+        "gateway_queue_write_count": 0,
+        "bot_api_call_count": 0,
+    }
+    delivery_mismatches = _mismatches(delivery, delivery_expected)
+    if delivery_mismatches:
+        raise ValueError(f"{label}: prior_delivery_approval mismatch: {delivery_mismatches}")
+    if not delivery.get("readiness_note"):
+        raise ValueError(f"{label}: prior_delivery_approval.readiness_note is required")
+
+
+def _require_truth(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("fixture_approval_truth") or {}
+    required_true = {
+        "p2_110_implementation_review_loaded",
+        "p2_111_delivery_approval_loaded",
+        "fixture_approval_package_ready",
+        "adapter_contract_ready",
+        "verifier_fixture_ready",
+        "blocker_mapping_ready",
+        "owner_review_required_before_readback",
+    }
+    missing = sorted(field for field in required_true if truth.get(field) is not True)
+    if missing:
+        raise ValueError(f"{label}: fixture approval ready flags must remain true: {missing}")
+
+    required_false = {
+        "canonical_runtime_target_read_enabled",
+        "live_query_enabled",
+        "runtime_readback_execution_enabled",
+        "gateway_queue_write_enabled",
+        "telegram_send_enabled",
+        "bot_api_call_enabled",
+        "report_receipt_write_enabled",
+        "result_capture_write_enabled",
+        "production_write_enabled",
+        "secret_read_enabled",
+        "destructive_operation_enabled",
+    }
+    unsafe = sorted(field for field in required_false if truth.get(field) is not False)
+    if unsafe:
+        raise ValueError(f"{label}: live read/send/write flags must remain false: {unsafe}")
+
+    zero_counts = {
+        "owner_approval_received_count",
+        "fixture_readback_execution_count",
+        "canonical_runtime_target_read_count",
+        "live_query_count",
+        "gateway_queue_write_count",
+        "telegram_send_count",
+        "bot_api_call_count",
+        "report_receipt_write_count",
+        "result_capture_write_count",
+        "production_write_count",
+    }
+    non_zero = sorted(field for field in zero_counts if truth.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: fixture approval live counters must remain zero: {non_zero}")
+    if not truth.get("truth_note"):
+        raise ValueError(f"{label}: fixture_approval_truth.truth_note is required")
+
+
+def _require_cards(payload: dict[str, Any], label: str) -> None:
+    cards = payload.get("fixture_approval_cards") or []
+    required = {
+        "report_delivery_fixture_readback",
+        "runtime_implementation_fixture_readback",
+        "telegram_failure_receipt_fixture_readback",
+        "reviewer_queue_fixture_preview",
+        "result_capture_fixture_link",
+    }
+    card_ids = {card.get("card_id") for card in cards}
+    if card_ids != required:
+        raise ValueError(f"{label}: fixture approval cards must match {sorted(required)}")
+
+    for card in cards:
+        card_id = card.get("card_id")
+        if card.get("owner_approval_required") is not True or card.get("fixture_only") is not True:
+            raise ValueError(f"{label}: card {card_id} must require owner approval and fixture_only")
+        if card.get("status") not in {"approval_required", "ready_for_owner_review", "blocked_by_policy"}:
+            raise ValueError(f"{label}: card {card_id} status is invalid")
+        if card.get("risk_tier") not in {"medium", "high", "critical"}:
+            raise ValueError(f"{label}: card {card_id} risk_tier is invalid")
+        if not card.get("required_fixture_fields") or not card.get("blocked_runtime_actions"):
+            raise ValueError(f"{label}: card {card_id} must list fixture fields and blocked actions")
+        if not _is_redacted_sha256(card.get("evidence_hash")):
+            raise ValueError(f"{label}: card {card_id} must expose redacted evidence_hash")
+
+
+def _require_contracts(payload: dict[str, Any], label: str) -> None:
+    contracts = payload.get("adapter_contracts") or []
+    required = {
+        "report_delivery_payload_to_fixture_readback",
+        "implementation_review_to_adapter_check",
+        "failure_receipt_to_fixture_verifier",
+        "result_capture_to_fixture_promotion",
+    }
+    contract_ids = {contract.get("contract_id") for contract in contracts}
+    if contract_ids != required:
+        raise ValueError(f"{label}: adapter contracts must match {sorted(required)}")
+
+    for contract in contracts:
+        contract_id = contract.get("contract_id")
+        if contract.get("status") not in {"ready", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: contract {contract_id} status is invalid")
+        if not contract.get("input_schema") or not contract.get("output_schema"):
+            raise ValueError(f"{label}: contract {contract_id} input/output schema is required")
+        if contract.get("canonical_target_read_enabled") is not False or contract.get("live_query_enabled") is not False:
+            raise ValueError(f"{label}: contract {contract_id} must not enable live target read/query")
+        if not contract.get("required_evidence"):
+            raise ValueError(f"{label}: contract {contract_id} required_evidence is required")
+        if not _is_redacted_sha256(contract.get("evidence_hash")):
+            raise ValueError(f"{label}: contract {contract_id} must expose redacted evidence_hash")
+
+
+def _require_checks(payload: dict[str, Any], label: str) -> None:
+    checks = payload.get("verifier_fixture_checks") or []
+    required = {
+        "fixture_payload_shape",
+        "no_live_target_reference",
+        "route_lock_fixture",
+        "redaction_fixture",
+        "result_capture_no_write_fixture",
+    }
+    check_ids = {check.get("check_id") for check in checks}
+    if check_ids != required:
+        raise ValueError(f"{label}: verifier fixture checks must match {sorted(required)}")
+
+    for check in checks:
+        check_id = check.get("check_id")
+        if check.get("status") not in {"ready", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: verifier check {check_id} status is invalid")
+        if check.get("live_verifier_enabled") is not False:
+            raise ValueError(f"{label}: verifier check {check_id} must not enable live verifier")
+        if not check.get("required_fixture") or not check.get("failure_if_missing"):
+            raise ValueError(f"{label}: verifier check {check_id} must include fixture and failure text")
+        if not _is_redacted_sha256(check.get("evidence_hash")):
+            raise ValueError(f"{label}: verifier check {check_id} must expose redacted evidence_hash")
+
+
+def _require_blockers(payload: dict[str, Any], label: str) -> None:
+    blockers = payload.get("blocker_mappings") or []
+    required = {
+        "canonical_runtime_target_blocked",
+        "gateway_queue_write_blocked",
+        "telegram_bot_send_blocked",
+        "report_receipt_write_blocked",
+        "result_capture_write_blocked",
+    }
+    blocker_ids = {blocker.get("blocker_id") for blocker in blockers}
+    if blocker_ids != required:
+        raise ValueError(f"{label}: blocker mappings must match {sorted(required)}")
+
+    for blocker in blockers:
+        blocker_id = blocker.get("blocker_id")
+        if blocker.get("severity") not in {"medium", "high", "critical"}:
+            raise ValueError(f"{label}: blocker {blocker_id} severity is invalid")
+        if blocker.get("status") not in {"mapped", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: blocker {blocker_id} status is invalid")
+        if not blocker.get("blocked_action") or not blocker.get("blocked_until"):
+            raise ValueError(f"{label}: blocker {blocker_id} blocked action/until is required")
+        if not _is_redacted_sha256(blocker.get("evidence_hash")):
+            raise ValueError(f"{label}: blocker {blocker_id} must expose redacted evidence_hash")
+
+
+def _require_actions(payload: dict[str, Any], label: str) -> None:
+    actions = payload.get("operator_actions") or []
+    required = {
+        "review_fixture_approval_cards",
+        "compare_adapter_contracts",
+        "confirm_no_live_query",
+        "reject_canonical_target_scope",
+        "promote_to_p2_113_fixture_readback",
+    }
+    action_ids = {action.get("action_id") for action in actions}
+    if action_ids != required:
+        raise ValueError(f"{label}: operator actions must match {sorted(required)}")
+
+    valid_types = {
+        "review_fixture_approval",
+        "compare_adapter_contract",
+        "confirm_no_live_query",
+        "reject_canonical_target",
+        "promote_to_p2_113",
+    }
+    for action in actions:
+        action_id = action.get("action_id")
+        if action.get("action_type") not in valid_types:
+            raise ValueError(f"{label}: action {action_id} action_type is invalid")
+        if action.get("runtime_readback_allowed") is not False:
+            raise ValueError(f"{label}: action {action_id} must not allow runtime readback")
+        if not action.get("operator_instruction"):
+            raise ValueError(f"{label}: action {action_id} operator_instruction is required")
+
+
+def _require_display_redaction(payload: dict[str, Any], label: str) -> None:
+    contract = payload.get("display_redaction_contract") or {}
+    if contract.get("redaction_required") is not True:
+        raise ValueError(f"{label}: display redaction must be required")
+    false_fields = {
+        "raw_prompt_display_allowed",
+        "private_reasoning_display_allowed",
+        "secret_value_display_allowed",
+        "raw_runtime_payload_display_allowed",
+        "internal_collaboration_content_display_allowed",
+    }
+    unsafe = sorted(field for field in false_fields if contract.get(field) is not False)
+    if unsafe:
+        raise ValueError(f"{label}: display redaction flags must remain false: {unsafe}")
+    if not contract.get("frontend_display_policy"):
+        raise ValueError(f"{label}: frontend_display_policy is required")
+
+
+def _require_no_forbidden_display_terms(payload: dict[str, Any], label: str) -> None:
+    serialized = json.dumps(payload, ensure_ascii=False).lower()
+    forbidden = {
+        "work_window_transcript",
+        "session_id",
+        "browser_context",
+        "authorization_header",
+        "raw telegram payload",
+        "private reasoning",
+        "raw prompt",
+        "chain-of-thought",
+    }
+    hits = sorted(term for term in forbidden if term in serialized)
+    if hits:
+        raise ValueError(f"{label}: forbidden display terms leaked: {hits}")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    rollups = payload.get("rollups") or {}
+    expected_counts = {
+        "fixture_approval_card_count": len(payload.get("fixture_approval_cards") or []),
+        "adapter_contract_count": len(payload.get("adapter_contracts") or []),
+        "verifier_fixture_check_count": len(payload.get("verifier_fixture_checks") or []),
+        "blocker_mapping_count": len(payload.get("blocker_mappings") or []),
+        "operator_action_count": len(payload.get("operator_actions") or []),
+        "approval_required_card_count": sum(
+            1 for card in payload.get("fixture_approval_cards") or [] if card.get("status") == "approval_required"
+        ),
+        "blocked_card_count": sum(
+            1 for card in payload.get("fixture_approval_cards") or [] if card.get("status") == "blocked_by_policy"
+        ),
+        "blocked_contract_count": sum(
+            1 for contract in payload.get("adapter_contracts") or [] if contract.get("status") == "blocked_by_policy"
+        ),
+        "blocked_check_count": sum(
+            1 for check in payload.get("verifier_fixture_checks") or [] if check.get("status") == "blocked_by_policy"
+        ),
+    }
+    mismatches = _mismatches(rollups, expected_counts)
+    if mismatches:
+        raise ValueError(f"{label}: rollup counts mismatch: {mismatches}")
+
+    zero_rollups = {
+        "owner_approval_received_count",
+        "fixture_readback_execution_count",
+        "canonical_runtime_target_read_count",
+        "live_query_count",
+        "gateway_queue_write_count",
+        "telegram_send_count",
+        "bot_api_call_count",
+        "report_receipt_write_count",
+        "result_capture_write_count",
+        "production_write_count",
+        "secret_read_count",
+        "destructive_operation_count",
+    }
+    non_zero = sorted(field for field in zero_rollups if rollups.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: live/send/write rollups must remain zero: {non_zero}")
+
+
+def _mismatches(actual: dict[str, Any], expected: dict[str, Any]) -> dict[str, dict[str, Any]]:
+    return {
+        key: {"expected": expected_value, "actual": actual.get(key)}
+        for key, expected_value in expected.items()
+        if actual.get(key) != expected_value
+    }
+
+
+def _is_redacted_sha256(value: Any) -> bool:
+    if not isinstance(value, str):
+        return False
+    if not value.startswith("sha256:") or len(value) != len("sha256:") + 64:
+        return False
+    digest = value.split(":", 1)[1]
+    return all(char in "0123456789abcdef" for char in digest)