feat(governance): 新增 release decision readback

apps/api/src/services/ai_agent_result_capture_release_decision_readback.py

@@ -0,0 +1,345 @@
+"""
+AI Agent result capture release decision readback snapshot.
+
+Loads the latest committed P2-139 release decision readback. This module only
+reads committed evidence from P2-138 release decision hold; it never treats a
+hold as a passed decision, grants release authorization, writes reviewer /
+Gateway queues, sends Telegram messages, writes result capture, writes learning
+records, updates PlayBook trust, reads secrets, writes production targets, or
+performs destructive operations.
+"""
+
+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_readback_*.json"
+_SCHEMA_VERSION = "ai_agent_result_capture_release_decision_readback_v1"
+_RUNTIME_AUTHORITY = "result_capture_release_decision_readback_only_no_live_write"
+
+
+def load_latest_ai_agent_result_capture_release_decision_readback(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed release decision readback."""
+    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 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")
+
+    label = str(latest)
+    _require_schema(payload, label)
+    _require_prior(payload, label)
+    _require_truth(payload, label)
+    _require_readback_groups(payload, label)
+    _require_blocked_transitions(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": "P2-139",
+        "next_task_id": "P2-140",
+        "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_release_decision_hold") or {}
+    expected = {
+        "schema_version": "ai_agent_result_capture_release_decision_hold_v1",
+        "release_decision_hold_count": 5,
+        "owner_decision_hold_count": 5,
+        "verifier_decision_hold_count": 5,
+        "rollback_decision_hold_count": 5,
+        "maintenance_window_decision_hold_count": 5,
+        "live_apply_decision_hold_count": 5,
+        "blocked_decision_transition_count": 6,
+        "operator_action_count": 5,
+        "approval_required_total": 12,
+        "blocked_total": 12,
+    }
+    expected.update(_zero_count_expectations())
+    mismatches = _mismatches(prior, expected)
+    if mismatches:
+        raise ValueError(f"{label}: prior_release_decision_hold mismatch: {mismatches}")
+    if not prior.get("readiness_note"):
+        raise ValueError(f"{label}: prior_release_decision_hold.readiness_note is required")
+
+
+def _require_truth(payload: dict[str, Any], label: str) -> None:
+    truth = payload.get("release_decision_readback_truth") or {}
+    required_true = {
+        "p2_138_release_decision_hold_loaded",
+        "release_decision_readback_ready",
+        "owner_decision_still_required",
+        "verifier_decision_still_required",
+        "rollback_decision_still_required",
+        "maintenance_window_decision_still_required",
+        "live_apply_decision_still_required",
+        "redaction_review_still_required",
+        "release_decision_readback_only",
+    }
+    missing = sorted(field for field in required_true if truth.get(field) is not True)
+    if missing:
+        raise ValueError(f"{label}: release decision readback flags must remain true: {missing}")
+
+    required_false = {
+        "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}: release decision readback live/send/write flags must remain false: {unsafe}")
+
+    non_zero = sorted(field for field in _zero_count_expectations() if truth.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: release decision readback live counters must remain zero: {non_zero}")
+    if not truth.get("truth_note"):
+        raise ValueError(f"{label}: release_decision_readback_truth.truth_note is required")
+
+
+def _require_readback_groups(payload: dict[str, Any], label: str) -> None:
+    groups = {
+        "release_decision_readbacks": ("readback_id", "release_decision_readback", "release decision readbacks"),
+        "owner_decision_readbacks": ("readback_id", "owner_decision_readback", "owner decision readbacks"),
+        "verifier_decision_readbacks": ("readback_id", "verifier_decision_readback", "verifier decision readbacks"),
+        "rollback_decision_readbacks": ("readback_id", "rollback_decision_readback", "rollback decision readbacks"),
+        "maintenance_window_decision_readbacks": ("readback_id", "maintenance_window_decision_readback", "maintenance window decision readbacks"),
+        "live_apply_decision_readbacks": ("readback_id", "live_apply_decision_readback", "live apply decision readbacks"),
+    }
+    for field, (id_field, mode, group_name) in groups.items():
+        items = payload.get(field) or []
+        _require_ids(items, id_field, _expected_ids(mode), label, group_name)
+        for item in items:
+            item_id = item.get(id_field)
+            if item.get("readback_mode") != mode:
+                raise ValueError(f"{label}: {group_name} {item_id} mode is invalid")
+            if item.get("readback_status") not in {"readback_complete", "approval_required", "blocked_by_policy"}:
+                raise ValueError(f"{label}: {group_name} {item_id} readback_status is invalid")
+            if not item.get("readback_summary"):
+                raise ValueError(f"{label}: {group_name} {item_id}.readback_summary is required")
+            if item.get("release_decision_passed") is not False:
+                raise ValueError(f"{label}: {group_name} {item_id} must keep release_decision_passed=false")
+            if item.get("release_authorization_granted") is not False:
+                raise ValueError(f"{label}: {group_name} {item_id} must keep release_authorization_granted=false")
+            if item.get("runtime_write_allowed") is not False:
+                raise ValueError(f"{label}: {group_name} {item_id} must keep runtime_write_allowed=false")
+            if item.get("telegram_send_allowed") is not False:
+                raise ValueError(f"{label}: {group_name} {item_id} must keep telegram_send_allowed=false")
+            if "evidence_hash" in item and not _is_redacted_sha256(item.get("evidence_hash")):
+                raise ValueError(f"{label}: {group_name} {item_id}.evidence_hash must be redacted sha256")
+
+
+def _require_blocked_transitions(payload: dict[str, Any], label: str) -> None:
+    items = payload.get("blocked_readback_transitions") or []
+    _require_count(items, 6, label, "blocked readback transitions")
+    for item in items:
+        item_id = item.get("transition_id")
+        if item.get("runtime_write_allowed") is not False:
+            raise ValueError(f"{label}: blocked readback transition {item_id} must keep runtime_write_allowed=false")
+        if item.get("status") not in {"blocked_by_policy", "approval_required"}:
+            raise ValueError(f"{label}: blocked readback transition {item_id} status is invalid")
+        if not item.get("reason"):
+            raise ValueError(f"{label}: blocked readback transition {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 = item.get("action_id")
+        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 {}
+    expected = {
+        "release_decision_readback_count": len(payload.get("release_decision_readbacks") or []),
+        "owner_decision_readback_count": len(payload.get("owner_decision_readbacks") or []),
+        "verifier_decision_readback_count": len(payload.get("verifier_decision_readbacks") or []),
+        "rollback_decision_readback_count": len(payload.get("rollback_decision_readbacks") or []),
+        "maintenance_window_decision_readback_count": len(payload.get("maintenance_window_decision_readbacks") or []),
+        "live_apply_decision_readback_count": len(payload.get("live_apply_decision_readbacks") or []),
+        "blocked_readback_transition_count": len(payload.get("blocked_readback_transitions") or []),
+        "operator_action_count": len(payload.get("operator_actions") or []),
+    }
+    expected.update(_zero_count_expectations())
+    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",
+    }
+    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 _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 _expected_ids(mode: str) -> set[str]:
+    return {
+        f"{mode}_result_capture",
+        f"{mode}_learning",
+        f"{mode}_playbook_trust",
+        f"{mode}_reviewer_queue",
+        f"{mode}_gateway_queue",
+    }
+
+
+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
+    }