feat(governance): 新增 owner-approved execution rehearsal

apps/api/src/services/ai_agent_result_capture_owner_approved_execution_rehearsal.py

@@ -0,0 +1,407 @@
+"""
+AI Agent result capture owner-approved execution rehearsal snapshot.
+
+Loads the latest committed P2-126 owner-approved execution rehearsal package.
+This module validates committed evidence only; it never applies writers, executes
+live applies, writes receipts, writes result captures, writes learning records,
+updates PlayBook trust, writes reviewer / Gateway queues, sends Telegram
+messages, reads canonical runtime targets, 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_result_capture_owner_approved_execution_rehearsal_*.json"
+_SCHEMA_VERSION = "ai_agent_result_capture_owner_approved_execution_rehearsal_v1"
+_RUNTIME_AUTHORITY = "result_capture_owner_approved_execution_rehearsal_only_no_live_write"
+
+
+def load_latest_ai_agent_result_capture_owner_approved_execution_rehearsal(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed owner-approved execution rehearsal package."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no AI Agent result capture owner-approved execution rehearsal 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_execution_rehearsals(payload, label)
+    _require_apply_checks(payload, label)
+    _require_verifier_rehearsals(payload, label)
+    _require_rollback_drills(payload, label)
+    _require_blocked_applies(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": "P2-126",
+        "next_task_id": "P2-127",
+        "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_result_capture_owner_promotion_review") or {}
+    expected = {
+        "schema_version": "ai_agent_result_capture_owner_promotion_review_v1",
+        "owner_promotion_packet_count": 5,
+        "execution_gate_check_count": 5,
+        "rollback_owner_review_count": 5,
+        "blocked_execution_write_count": 6,
+        "operator_action_count": 5,
+        "owner_promotion_approved_count": 0,
+        "execution_gate_pass_count": 0,
+        "rollback_owner_confirmed_count": 0,
+        "post_write_verifier_ready_count": 0,
+        "writer_apply_count": 0,
+        "execution_apply_count": 0,
+        "receipt_write_count": 0,
+        "result_capture_write_count": 0,
+        "learning_write_count": 0,
+        "playbook_trust_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,
+    }
+    mismatches = _mismatches(prior, expected)
+    if mismatches:
+        raise ValueError(f"{label}: prior_result_capture_owner_promotion_review mismatch: {mismatches}")
+    if not prior.get("readiness_note"):
+        raise ValueError(f"{label}: prior_result_capture_owner_promotion_review.readiness_note is required")
+
+
+def _require_truth(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("execution_rehearsal_truth") or {}
+    required_true = {
+        "p2_125_owner_promotion_loaded",
+        "owner_approval_record_required",
+        "no_write_apply_rehearsal_ready",
+        "post_write_verifier_rehearsal_ready",
+        "rollback_drill_ready",
+        "idempotency_replay_required",
+        "redaction_review_required",
+        "execution_rehearsal_only",
+    }
+    missing = sorted(field for field in required_true if truth.get(field) is not True)
+    if missing:
+        raise ValueError(f"{label}: rehearsal ready flags must remain true: {missing}")
+    required_false = {
+        "owner_approval_record_received",
+        "no_write_apply_executed",
+        "post_write_verifier_passed",
+        "rollback_drill_confirmed",
+        "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}: live execution/send/write flags must remain false: {unsafe}")
+    zero_counts = {
+        "owner_approval_record_received_count",
+        "no_write_apply_execution_count",
+        "post_write_verifier_pass_count",
+        "rollback_drill_confirmed_count",
+        "writer_apply_count",
+        "execution_apply_count",
+        "receipt_write_count",
+        "reviewer_queue_write_count",
+        "gateway_queue_write_count",
+        "telegram_send_count",
+        "bot_api_call_count",
+        "report_receipt_write_count",
+        "result_capture_write_count",
+        "learning_write_count",
+        "playbook_trust_write_count",
+        "production_write_count",
+        "secret_read_count",
+        "destructive_operation_count",
+    }
+    non_zero = sorted(field for field in zero_counts if truth.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: rehearsal live counters must remain zero: {non_zero}")
+    if not truth.get("truth_note"):
+        raise ValueError(f"{label}: execution_rehearsal_truth.truth_note is required")
+
+
+def _require_execution_rehearsals(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("execution_rehearsals") or []
+    required = {
+        "rehearsal_result_capture_writer",
+        "rehearsal_learning_writer",
+        "rehearsal_playbook_trust_writer",
+        "rehearsal_reviewer_queue_writer",
+        "rehearsal_gateway_queue_writer",
+    }
+    ids = {item.get("rehearsal_id") for item in items}
+    if ids != required:
+        raise ValueError(f"{label}: execution rehearsals must match {sorted(required)}")
+    for item in items:
+        item_id = item.get("rehearsal_id")
+        if item.get("rehearsal_mode") != "no_write_apply_rehearsal":
+            raise ValueError(f"{label}: rehearsal {item_id} must remain no-write apply rehearsal")
+        if item.get("no_write_only") is not True or item.get("live_apply_enabled") is not False:
+            raise ValueError(f"{label}: rehearsal {item_id} must not enable live apply")
+        if item.get("status") not in {"ready_for_rehearsal_review", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: rehearsal {item_id} status is invalid")
+        if not item.get("expected_output") or not _is_redacted_sha256(item.get("evidence_hash")):
+            raise ValueError(f"{label}: rehearsal {item_id} must include expected output and redacted evidence")
+
+
+def _require_apply_checks(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("no_write_apply_checks") or []
+    required = {
+        "check_owner_approval_record",
+        "check_idempotency_replay",
+        "check_payload_redaction",
+        "check_post_write_verifier",
+        "check_production_write_block",
+    }
+    ids = {item.get("check_id") for item in items}
+    if ids != required:
+        raise ValueError(f"{label}: no-write apply checks must match {sorted(required)}")
+    for item in items:
+        item_id = item.get("check_id")
+        if item.get("check_mode") != "no_write_apply_gate":
+            raise ValueError(f"{label}: check {item_id} must remain no-write apply gate")
+        if item.get("required_before_apply") is not True or item.get("live_apply_enabled") is not False:
+            raise ValueError(f"{label}: check {item_id} must not enable live apply")
+        if item.get("status") not in {"ready_for_rehearsal_review", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: check {item_id} status is invalid")
+        if not item.get("failure_if_missing"):
+            raise ValueError(f"{label}: check {item_id} failure_if_missing is required")
+
+
+def _require_verifier_rehearsals(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("post_write_verifier_rehearsals") or []
+    required = {
+        "verifier_result_capture",
+        "verifier_learning",
+        "verifier_playbook_trust",
+        "verifier_reviewer_queue",
+        "verifier_gateway_queue",
+    }
+    ids = {item.get("verifier_id") for item in items}
+    if ids != required:
+        raise ValueError(f"{label}: verifier rehearsals must match {sorted(required)}")
+    for item in items:
+        item_id = item.get("verifier_id")
+        if item.get("verifier_mode") != "fixture_only_no_live_read":
+            raise ValueError(f"{label}: verifier {item_id} must remain fixture-only")
+        if item.get("verifier_ready") is not False or item.get("live_read_enabled") is not False:
+            raise ValueError(f"{label}: verifier {item_id} must not enable live read")
+        if item.get("status") not in {"ready_for_rehearsal_review", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: verifier {item_id} status is invalid")
+        if not item.get("expected_signal"):
+            raise ValueError(f"{label}: verifier {item_id} expected_signal is required")
+
+
+def _require_rollback_drills(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("rollback_drills") or []
+    required = {
+        "rollback_drill_result_capture",
+        "rollback_drill_learning",
+        "rollback_drill_playbook_trust",
+        "rollback_drill_reviewer_queue",
+        "rollback_drill_gateway_queue",
+    }
+    ids = {item.get("drill_id") for item in items}
+    if ids != required:
+        raise ValueError(f"{label}: rollback drills must match {sorted(required)}")
+    for item in items:
+        item_id = item.get("drill_id")
+        if item.get("rollback_owner_required") is not True or item.get("rollback_drill_confirmed") is not False:
+            raise ValueError(f"{label}: rollback drill {item_id} must require unconfirmed owner")
+        if item.get("status") not in {"ready_for_rehearsal_review", "approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: rollback drill {item_id} status is invalid")
+        if not item.get("drill_note"):
+            raise ValueError(f"{label}: rollback drill {item_id} drill_note is required")
+
+
+def _require_blocked_applies(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("blocked_live_applies") or []
+    required = {
+        "blocked_writer_apply",
+        "blocked_execution_apply",
+        "blocked_receipt_write",
+        "blocked_result_capture_write",
+        "blocked_gateway_queue_write",
+        "blocked_telegram_send",
+    }
+    ids = {item.get("blocker_id") for item in items}
+    if ids != required:
+        raise ValueError(f"{label}: blocked live applies must match {sorted(required)}")
+    for item in items:
+        item_id = item.get("blocker_id")
+        if item.get("status") not in {"approval_required", "blocked_by_policy"}:
+            raise ValueError(f"{label}: blocker {item_id} status is invalid")
+        if item.get("severity") not in {"high", "critical"}:
+            raise ValueError(f"{label}: blocker {item_id} severity is invalid")
+        if not item.get("blocked_action") or not item.get("blocked_until"):
+            raise ValueError(f"{label}: blocker {item_id} must include blocked_action and blocked_until")
+        if not _is_redacted_sha256(item.get("evidence_hash")):
+            raise ValueError(f"{label}: blocker {item_id} must expose redacted evidence_hash")
+
+
+def _require_actions(payload: dict[str, Any], label: str) -> None:
+    actions = payload.get("operator_actions") or []
+    required = {
+        "review_execution_rehearsals",
+        "verify_no_write_apply_checks",
+        "inspect_post_write_verifier_rehearsal",
+        "confirm_rollback_drills",
+        "hold_live_apply_gate",
+    }
+    ids = {action.get("action_id") for action in actions}
+    if ids != required:
+        raise ValueError(f"{label}: operator actions must match {sorted(required)}")
+    for action in actions:
+        action_id = action.get("action_id")
+        if action.get("runtime_write_allowed") is not False:
+            raise ValueError(f"{label}: action {action_id} must not allow runtime write")
+        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 {}
+    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 {}
+    rehearsals = payload.get("execution_rehearsals") or []
+    checks = payload.get("no_write_apply_checks") or []
+    verifiers = payload.get("post_write_verifier_rehearsals") or []
+    drills = payload.get("rollback_drills") or []
+    blockers = payload.get("blocked_live_applies") or []
+    actions = payload.get("operator_actions") or []
+    expected = {
+        "execution_rehearsal_count": len(rehearsals),
+        "no_write_apply_check_count": len(checks),
+        "post_write_verifier_rehearsal_count": len(verifiers),
+        "rollback_drill_count": len(drills),
+        "blocked_live_apply_count": len(blockers),
+        "operator_action_count": len(actions),
+        "approval_required_rehearsal_count": sum(1 for item in rehearsals if item.get("status") == "approval_required"),
+        "blocked_rehearsal_count": sum(1 for item in rehearsals if item.get("status") == "blocked_by_policy"),
+        "approval_required_check_count": sum(1 for item in checks if item.get("status") == "approval_required"),
+        "blocked_check_count": sum(1 for item in checks if item.get("status") == "blocked_by_policy"),
+        "approval_required_verifier_count": sum(1 for item in verifiers if item.get("status") == "approval_required"),
+        "blocked_verifier_count": sum(1 for item in verifiers if item.get("status") == "blocked_by_policy"),
+        "approval_required_rollback_count": sum(1 for item in drills if item.get("status") == "approval_required"),
+        "blocked_rollback_count": sum(1 for item in drills if item.get("status") == "blocked_by_policy"),
+        "critical_blocker_count": sum(1 for item in blockers if item.get("severity") == "critical"),
+        "owner_approval_record_received_count": 0,
+        "no_write_apply_execution_count": 0,
+        "post_write_verifier_pass_count": 0,
+        "rollback_drill_confirmed_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,
+    }
+    mismatches = _mismatches(rollups, expected)
+    if mismatches:
+        raise ValueError(f"{label}: rollup counts mismatch: {mismatches}")
+
+
+def _require_no_forbidden_display_terms(payload: dict[str, Any], label: str) -> None:
+    serialized = json.dumps(payload, ensure_ascii=False)
+    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 present: {hits}")
+
+
+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)
+
+
+def _mismatches(payload: dict[str, Any], expected: dict[str, Any]) -> dict[str, dict[str, Any]]:
+    return {
+        key: {"expected": value, "actual": payload.get(key)}
+        for key, value in expected.items()
+        if payload.get(key) != value
+    }