feat(governance): 新增 P2-143 owner response 預檢

2026-06-14 14:37:04 +08:00
parent 1a2c9e36c7
commit 755b0a8d30
12 changed files with 1366 additions and 7 deletions
--- a/apps/api/src/services/ai_agent_result_capture_release_decision_owner_response_preflight.py
+++ b/apps/api/src/services/ai_agent_result_capture_release_decision_owner_response_preflight.py
@@ -0,0 +1,376 @@
+"""
+AI Agent result capture release decision owner-response preflight snapshot.
+
+Loads the latest committed P2-143 owner-response preflight package. This module
+only turns P2-141 decision input prep into read-only intake checks and rejection
+guards; it never treats preflight as receipt, approval, reviewer queue write,
+Gateway queue write, Telegram send, Bot API call, result capture, learning,
+PlayBook trust, secret read, production write, or destructive operation.
+"""
+
+from __future__ import annotations
+
+import json
+import re
+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_result_capture_release_decision_owner_response_preflight_*.json"
+_SCHEMA_VERSION = "ai_agent_result_capture_release_decision_owner_response_preflight_v1"
+_RUNTIME_AUTHORITY = "result_capture_release_decision_owner_response_preflight_only_no_live_write"
+
+
+def load_latest_ai_agent_result_capture_release_decision_owner_response_preflight(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed release decision owner-response preflight."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(
+            f"no AI Agent result capture release decision owner response preflight 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_response_intake_lanes(payload, label)
+    _require_validation_checks(payload, label)
+    _require_rejection_guards(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": "P2-143",
+        "next_task_id": "P2-144",
+        "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:
+    prior = payload.get("prior_decision_input_prep") or {}
+    expected = {
+        "schema_version": "ai_agent_result_capture_release_decision_input_prep_v1",
+        "current_task_id": "P2-141",
+        "next_task_id": "P2-142",
+        "decision_input_packet_count": 5,
+        "missing_input_field_count": 18,
+        "blocked_input_transition_count": 6,
+        "operator_action_count": 5,
+        "approval_required_subtotal": 12,
+        "blocked_and_critical_subtotal": 12,
+    }
+    expected.update(_zero_count_expectations())
+    mismatches = _mismatches(prior, expected)
+    if mismatches:
+        raise ValueError(f"{label}: prior_decision_input_prep mismatch: {mismatches}")
+    if not prior.get("readiness_note"):
+        raise ValueError(f"{label}: prior_decision_input_prep.readiness_note is required")
+
+
+def _require_truth(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("owner_response_preflight_truth") or {}
+    required_true = {
+        "p2_141_input_prep_loaded",
+        "p2_142_war_room_baseline_preserved",
+        "required_owner_fields_declared",
+        "required_verifier_fields_declared",
+        "required_rollback_fields_declared",
+        "required_maintenance_window_fields_declared",
+        "required_live_apply_fields_declared",
+        "redaction_contract_loaded",
+        "preflight_only_mode",
+        "no_raw_payload_required",
+        "rejection_policy_active",
+    }
+    missing = sorted(field for field in required_true if truth.get(field) is not True)
+    if missing:
+        raise ValueError(f"{label}: owner response preflight flags must remain true: {missing}")
+
+    required_false = {
+        "owner_response_received",
+        "owner_response_accepted",
+        "owner_response_rejected",
+        "redacted_payload_ingested",
+        "owner_release_authorized",
+        "owner_release_approved",
+        "owner_review_approved",
+        "owner_decision_approved",
+        "verifier_decision_approved",
+        "maintenance_window_approved",
+        "rollback_owner_confirmed",
+        "post_release_verifier_ready",
+        "final_release_candidate_approved",
+        "final_release_candidate_passed",
+        "release_decision_passed",
+        "release_authorization_granted",
+        "release_authorization_passed",
+        "rollback_release_passed",
+        "live_apply_release_passed",
+        "writer_apply_enabled",
+        "execution_apply_enabled",
+        "receipt_write_enabled",
+        "reviewer_queue_write_enabled",
+        "gateway_queue_write_enabled",
+        "telegram_send_enabled",
+        "bot_api_call_enabled",
+        "report_receipt_write_enabled",
+        "result_capture_write_enabled",
+        "learning_write_enabled",
+        "playbook_trust_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}: owner response live/send/write flags must remain false: {unsafe}")
+
+    non_zero = sorted(field for field in _owner_response_zero_counts() if truth.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: owner response live counters must remain zero: {non_zero}")
+    if not truth.get("truth_note"):
+        raise ValueError(f"{label}: owner_response_preflight_truth.truth_note is required")
+
+
+def _require_response_intake_lanes(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("response_intake_lanes") or []
+    _require_ids(
+        items,
+        "lane_id",
+        {
+            "owner_release_response_preflight",
+            "verifier_response_preflight",
+            "rollback_owner_response_preflight",
+            "maintenance_window_response_preflight",
+            "live_apply_response_preflight",
+        },
+        label,
+        "response intake lanes",
+    )
+    for item in items:
+        item_id = item.get("lane_id")
+        expected = {
+            "current_task_id": "P2-143",
+            "target_next_task_id": "P2-144",
+            "intake_status": "waiting_for_external_owner_response",
+            "accepted": False,
+            "redacted_payload_ingested": False,
+            "runtime_write_allowed": False,
+            "telegram_send_allowed": False,
+        }
+        mismatches = _mismatches(item, expected)
+        if mismatches:
+            raise ValueError(f"{label}: response intake lane {item_id} mismatch: {mismatches}")
+        required_fields = item.get("required_fields") or []
+        if not isinstance(required_fields, list) or not required_fields:
+            raise ValueError(f"{label}: response intake lane {item_id}.required_fields is required")
+        if not item.get("source_packet_id"):
+            raise ValueError(f"{label}: response intake lane {item_id}.source_packet_id is required")
+        if not item.get("preflight_summary"):
+            raise ValueError(f"{label}: response intake lane {item_id}.preflight_summary is required")
+        if not _is_redacted_sha256(item.get("evidence_hash")):
+            raise ValueError(f"{label}: response intake lane {item_id}.evidence_hash must be redacted sha256")
+
+
+def _require_validation_checks(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("intake_validation_checks") or []
+    _require_count(items, 6, label, "intake validation checks")
+    for item in items:
+        item_id = item.get("check_id")
+        if item.get("status") not in {"waiting_for_owner_response", "not_satisfied", "blocked_by_policy"}:
+            raise ValueError(f"{label}: intake validation check {item_id} status is invalid")
+        if item.get("runtime_write_allowed") is not False:
+            raise ValueError(f"{label}: intake validation check {item_id} must keep runtime_write_allowed=false")
+        if not item.get("requirement"):
+            raise ValueError(f"{label}: intake validation check {item_id}.requirement is required")
+
+
+def _require_rejection_guards(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("rejection_guards") or []
+    _require_count(items, 6, label, "rejection guards")
+    for item in items:
+        item_id = item.get("guard_id")
+        if item.get("status") != "active":
+            raise ValueError(f"{label}: rejection guard {item_id} status must remain active")
+        if item.get("runtime_write_allowed") is not False:
+            raise ValueError(f"{label}: rejection guard {item_id} must keep runtime_write_allowed=false")
+        if not item.get("reason"):
+            raise ValueError(f"{label}: rejection guard {item_id}.reason is required")
+
+
+def _require_actions(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("operator_actions") or []
+    _require_count(items, 5, label, "operator actions")
+    for item in items:
+        item_id = str(item.get("action_id"))
+        if "p2_139" in item_id or "p2_140" in item_id or "p2_141" in item_id:
+            raise ValueError(f"{label}: operator action {item_id} must not point back to P2-139/P2-140/P2-141")
+        if item.get("runtime_write_allowed") is not False:
+            raise ValueError(f"{label}: operator action {item_id} must keep runtime_write_allowed=false")
+        if item.get("status") not in {"ready_for_operator_review", "approval_required"}:
+            raise ValueError(f"{label}: operator action {item_id} status is invalid")
+        if not item.get("operator_instruction"):
+            raise ValueError(f"{label}: operator action {item_id}.operator_instruction is required")
+
+
+def _require_display_redaction(payload: dict[str, Any], label: str) -> None:
+    contract = payload.get("display_redaction_contract") or {}
+    expected = {
+        "redaction_required": True,
+        "raw_prompt_display_allowed": False,
+        "private_reasoning_display_allowed": False,
+        "secret_value_display_allowed": False,
+        "raw_runtime_payload_display_allowed": False,
+        "internal_collaboration_content_display_allowed": False,
+    }
+    mismatches = _mismatches(contract, expected)
+    if mismatches:
+        raise ValueError(f"{label}: display_redaction_contract mismatch: {mismatches}")
+    if not contract.get("frontend_display_policy"):
+        raise ValueError(f"{label}: display_redaction_contract.frontend_display_policy is required")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    rollups = payload.get("rollups") or {}
+    required_owner_field_count = sum(len(item.get("required_fields") or []) for item in payload.get("response_intake_lanes") or [])
+    expected = {
+        "response_intake_lane_count": len(payload.get("response_intake_lanes") or []),
+        "required_owner_field_count": required_owner_field_count,
+        "intake_validation_check_count": len(payload.get("intake_validation_checks") or []),
+        "rejection_guard_count": len(payload.get("rejection_guards") or []),
+        "operator_action_count": len(payload.get("operator_actions") or []),
+        "waiting_external_response_count": len(payload.get("response_intake_lanes") or []),
+        "approval_required_subtotal": 12,
+        "blocked_and_critical_subtotal": 12,
+    }
+    expected.update(_owner_response_zero_counts())
+    mismatches = _mismatches(rollups, expected)
+    if mismatches:
+        raise ValueError(f"{label}: rollups mismatch: {mismatches}")
+
+
+def _require_no_forbidden_display_terms(payload: Any, label: str) -> None:
+    forbidden = {
+        "work_window_transcript",
+        "internal_collaboration_transcript",
+        "raw_prompt",
+        "private_reasoning",
+        "chain_of_thought",
+        "authorization_header",
+        "secret_value",
+        "工作視窗",
+        "內部協作",
+        "原始提示詞",
+        "私有推理",
+        "原始 runtime payload",
+        "raw runtime payload",
+        "authorization header",
+    }
+    found = sorted(term for term in forbidden if _contains_term(payload, term))
+    if found:
+        raise ValueError(f"{label}: forbidden display terms leaked: {found}")
+
+
+def _contains_term(value: Any, term: str) -> bool:
+    if isinstance(value, str):
+        return term.lower() in value.lower()
+    if isinstance(value, dict):
+        return any(_contains_term(child, term) for child in value.values())
+    if isinstance(value, list):
+        return any(_contains_term(child, term) for child in value)
+    return False
+
+
+def _owner_response_zero_counts() -> dict[str, int]:
+    return {
+        "owner_response_received_count": 0,
+        "owner_response_accepted_count": 0,
+        "owner_response_rejected_count": 0,
+        "redacted_payload_ingested_count": 0,
+        **_zero_count_expectations(),
+    }
+
+
+def _zero_count_expectations() -> dict[str, int]:
+    return {
+        "owner_release_authorized_count": 0,
+        "owner_release_approved_count": 0,
+        "owner_review_approved_count": 0,
+        "owner_decision_approved_count": 0,
+        "verifier_decision_approved_count": 0,
+        "maintenance_window_approved_count": 0,
+        "rollback_owner_confirmed_count": 0,
+        "post_release_verifier_ready_count": 0,
+        "final_release_candidate_approved_count": 0,
+        "final_release_candidate_pass_count": 0,
+        "release_decision_pass_count": 0,
+        "release_authorization_granted_count": 0,
+        "release_authorization_pass_count": 0,
+        "rollback_release_pass_count": 0,
+        "live_apply_release_pass_count": 0,
+        "writer_apply_count": 0,
+        "execution_apply_count": 0,
+        "receipt_write_count": 0,
+        "reviewer_queue_write_count": 0,
+        "gateway_queue_write_count": 0,
+        "telegram_send_count": 0,
+        "bot_api_call_count": 0,
+        "report_receipt_write_count": 0,
+        "result_capture_write_count": 0,
+        "learning_write_count": 0,
+        "playbook_trust_write_count": 0,
+        "production_write_count": 0,
+        "secret_read_count": 0,
+        "destructive_operation_count": 0,
+    }
+
+
+def _require_ids(items: list[dict[str, Any]], field: str, expected_ids: set[str], label: str, group_name: str) -> None:
+    actual_ids = {str(item.get(field)) for item in items}
+    if actual_ids != expected_ids:
+        raise ValueError(f"{label}: {group_name} ids mismatch: expected={sorted(expected_ids)} actual={sorted(actual_ids)}")
+
+
+def _require_count(items: list[Any], expected: int, label: str, group_name: str) -> None:
+    if len(items) != expected:
+        raise ValueError(f"{label}: expected {expected} {group_name}, got {len(items)}")
+
+
+def _is_redacted_sha256(value: Any) -> bool:
+    return isinstance(value, str) and re.fullmatch(r"sha256:[0-9a-f]{64}", value) is not None
+
+
+def _mismatches(payload: dict[str, Any], expected: dict[str, Any]) -> dict[str, dict[str, Any]]:
+    return {
+        key: {"expected": expected_value, "actual": payload.get(key)}
+        for key, expected_value in expected.items()
+        if payload.get(key) != expected_value
+    }