wooo/awoooi
1
1
Fork 0
Code Issues 2 Pull Requests 1 Actions 2 Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'gitea/main' into codex/delivery-workbench-release-20260626-ffsync

Browse Source
This commit is contained in:
ogt
2026-06-27 00:05:47 +08:00
parent 47797d9684 b7045a412c
commit 3f13d50ff5
11 changed files with 659 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
apps/api/src/api/v1/agents.py
View File

@@ -1115,18 +1115,18 @@ async def get_agent_low_medium_risk_whitelist() -> dict[str, Any]:
@router.get(
"/agent-high-risk-owner-review-queue",
response_model=dict[str, Any],
summary="取得 P2-409 AI Agent 高風險 Owner Review Queue",
summary="取得 P2-409 AI Agent 高風險受控執行 / Break-glass 佇列",
description=(
"讀取最新已提交的 P2-409 AI Agent 高風險 Owner Review Queue 只讀快照；"
"此端點只呈現 high / critical action queue、approval packet、rejection guard、"
"reviewer checklist、Telegram policy 與高風險暫停邊界。"
"它不啟動 auto worker、不執行 live action、不寫 Gateway queue、不送 Telegram、"
"讀取最新已提交的 P2-409 AI Agent 高風險受控執行 / critical break-glass 只讀快照；"
"此端點呈現 high controlled apply queue、critical break-glass queue、packet、"
"rejection guard、reviewer checklist、Telegram policy 與執行邊界。"
"它不自行啟動 auto worker、不執行 live action、不寫 Gateway queue、不送 Telegram、"
"不呼叫 Bot API、不寫 receipt production target、不寫 production、不讀 secret、"
"不呼叫付費 API、不改主機、不執行 kubectl 或不可逆操作。"
),
)
async def get_agent_high_risk_owner_review_queue() -> dict[str, Any]:
"""回傳最新 P2-409 high-risk owner review queue 只讀快照。"""
"""回傳最新 P2-409 high-risk controlled apply queue 只讀快照。"""
try:
payload = await asyncio.to_thread(load_latest_ai_agent_high_risk_owner_review_queue)
return redact_public_lan_topology(payload)
@@ -1139,7 +1139,7 @@ async def get_agent_high_risk_owner_review_queue() -> dict[str, Any]:
logger.error("ai_agent_high_risk_owner_review_queue_invalid", error=str(exc))
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail="P2-409 AI Agent 高風險 Owner Review Queue 快照無效",
detail="P2-409 AI Agent 高風險受控執行 / Break-glass 佇列快照無效",
) from exc

56
apps/api/src/services/ai_agent_high_risk_owner_review_queue.py
View File

@@ -1,10 +1,11 @@
"""
P2-409 AI Agent high-risk owner review queue snapshot.
P2-409 AI Agent controlled-apply / break-glass queue snapshot.
Loads the latest committed high-risk owner review queue. This module only
validates read-only approval packets, rejection guards, and reviewer checklists.
It does not run workers, send Telegram, write Gateway queues, read secrets, call
paid APIs, mutate hosts, run kubectl, or write production state.
Loads the latest committed controlled-apply / critical break-glass queue. This
module validates read-only controlled apply packets, rejection guards, and
reviewer checklists. It does not run workers, send Telegram, write Gateway
queues, read secrets, call paid APIs, mutate hosts, run kubectl, or write
production state.
"""
from __future__ import annotations
@@ -18,7 +19,7 @@ from src.services.snapshot_paths import default_evaluations_dir
_DEFAULT_EVALUATIONS_DIR = default_evaluations_dir(Path(__file__))
_SNAPSHOT_PATTERN = "ai_agent_high_risk_owner_review_queue_*.json"
_SCHEMA_VERSION = "ai_agent_high_risk_owner_review_queue_v1"
_RUNTIME_AUTHORITY = "high_risk_owner_review_queue_no_live_execution_committed_snapshot"
_RUNTIME_AUTHORITY = "controlled_apply_break_glass_queue_readback_no_live_execution"
_EXPECTED_CURRENT_TASK = "P2-409"
_EXPECTED_NEXT_TASK = "P2-410"
_EXPECTED_CANONICAL_ROOM = "AwoooI SRE 戰情室"
@@ -38,13 +39,15 @@ _TRUE_TRUTH_FLAGS = {
"p2_110e_work_items_owner_review_loaded",
"telegram_egress_inventory_loaded",
"telegram_owner_request_draft_loaded",
"all_high_risk_actions_paused",
"approval_packets_ready",
"rejection_guards_ready",
"reviewer_checklists_ready",
"high_risk_owner_review_required",
"high_risk_controlled_apply_enabled",
"critical_break_glass_required",
}
_FALSE_TRUTH_FLAGS = {
"all_high_risk_actions_paused",
"high_risk_owner_review_required",
"auto_worker_enabled",
"live_execution_enabled",
"gateway_queue_write_enabled",
@@ -77,12 +80,14 @@ _ZERO_TRUTH_COUNTS = {
"redacted_payload_ingested_count_24h",
}
_TRUE_BOUNDARY_FLAGS = {
"read_only_owner_review_queue_allowed",
"controlled_apply_queue_readback_allowed",
"critical_break_glass_queue_readback_allowed",
"approval_packet_preview_allowed",
"rejection_guard_preview_allowed",
"reviewer_checklist_allowed",
}
_FALSE_BOUNDARY_FLAGS = {
"read_only_owner_review_queue_allowed",
"auto_worker_enabled",
"live_execution_enabled",
"gateway_queue_write_enabled",
@@ -133,11 +138,11 @@ _FORBIDDEN_PUBLIC_TERMS = {
def load_latest_ai_agent_high_risk_owner_review_queue(
evaluations_dir: Path | None = None,
) -> dict[str, Any]:
"""Load the newest committed P2-409 high-risk owner review queue snapshot."""
"""Load the newest committed P2-409 controlled apply queue snapshot."""
directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
if not candidates:
raise FileNotFoundError(f"no AI Agent high-risk owner review queue snapshots found in {directory}")
raise FileNotFoundError(f"no AI Agent controlled apply queue snapshots found in {directory}")
latest = candidates[-1]
with latest.open(encoding="utf-8") as handle:
@@ -222,7 +227,8 @@ def _require_queue_items(payload: dict[str, Any], label: str) -> None:
if item.get("risk_tier") not in {"high", "critical"}:
raise ValueError(f"{label}: queue item {item_id} must be high or critical")
if item.get("queue_status") not in {
"paused_owner_review_required",
"controlled_apply_packet_ready",
"critical_break_glass_required",
"blocked_missing_owner_response",
"approval_packet_preview_ready",
}:
@@ -238,7 +244,13 @@ def _require_queue_items(payload: dict[str, Any], label: str) -> None:
):
if not item.get(field):
raise ValueError(f"{label}: queue item {item_id} missing {field}")
for flag in ("owner_response_required", "rollback_owner_required", "postcheck_required"):
expected_owner_response = item.get("risk_tier") == "critical"
if item.get("owner_response_required") is not expected_owner_response:
raise ValueError(
f"{label}: queue item {item_id}.owner_response_required must be "
f"{expected_owner_response}"
)
for flag in ("rollback_owner_required", "postcheck_required"):
if item.get(flag) is not True:
raise ValueError(f"{label}: queue item {item_id}.{flag} must remain true")
for flag in (
@@ -265,7 +277,11 @@ def _require_approval_packets(payload: dict[str, Any], label: str) -> None:
packet_id = packet.get("approval_packet_id") or "<missing>"
if packet.get("queue_item_id") not in queue_ids:
raise ValueError(f"{label}: approval packet {packet_id} references unknown queue item")
if packet.get("packet_status") not in {"draft_ready_owner_response_required", "blocked_missing_owner_response"}:
if packet.get("packet_status") not in {
"controlled_apply_packet_ready",
"break_glass_packet_ready",
"blocked_missing_owner_response",
}:
raise ValueError(f"{label}: approval packet {packet_id} status is invalid")
for field in ("required_owner_fields", "required_evidence_refs", "reviewer_checklist_id", "rejection_guard_ids"):
if not packet.get(field):
@@ -334,10 +350,10 @@ def _require_reviewer_checklists(payload: dict[str, Any], label: str) -> None:
def _require_routing_policy(payload: dict[str, Any], label: str) -> None:
policy = payload.get("routing_policy") or {}
expected = {
"high_risk_default_route": "pause_to_owner_review_queue",
"critical_risk_default_route": "pause_to_owner_review_queue",
"low_medium_runtime_route": "pause_until_owner_approved_runtime_gate",
"owner_response_required": True,
"high_risk_default_route": "controlled_apply_queue",
"critical_risk_default_route": "critical_break_glass_queue",
"low_medium_runtime_route": "controlled_apply_queue",
"owner_response_required": False,
"verbal_approval_accepted": False,
"redacted_payload_only": True,
}
@@ -414,6 +430,10 @@ def _require_rollups(payload: dict[str, Any], label: str) -> None:
"rollback_owner_required_count": sum(1 for item in items if item.get("rollback_owner_required") is True),
"postcheck_required_count": sum(1 for item in items if item.get("postcheck_required") is True),
"blocked_runtime_action_count": len(blocked_actions),
"controlled_apply_queue_count": sum(1 for item in items if item.get("risk_tier") == "high"),
"critical_break_glass_queue_count": sum(1 for item in items if item.get("risk_tier") == "critical"),
"owner_response_required_count": sum(1 for item in items if item.get("owner_response_required") is True),
"high_risk_owner_review_required_count": 0,
}
mismatches = {
key: {"expected": value, "actual": rollups.get(key)}

32
apps/api/tests/test_ai_agent_high_risk_owner_review_queue.py
View File

@@ -37,9 +37,21 @@ def test_load_latest_ai_agent_high_risk_owner_review_queue_reads_newest_file(tmp
assert loaded["program_status"]["current_task_id"] == "P2-409"
assert loaded["program_status"]["next_task_id"] == "P2-410"
assert loaded["program_status"]["read_only_mode"] is True
assert (
loaded["program_status"]["runtime_authority"]
== "controlled_apply_break_glass_queue_readback_no_live_execution"
)
assert loaded["queue_truth"]["all_high_risk_actions_paused"] is False
assert loaded["queue_truth"]["high_risk_owner_review_required"] is False
assert loaded["queue_truth"]["high_risk_controlled_apply_enabled"] is True
assert loaded["queue_truth"]["critical_break_glass_required"] is True
assert loaded["rollups"]["queue_item_count"] == 7
assert loaded["rollups"]["approval_packet_count"] == 7
assert loaded["rollups"]["rejection_guard_count"] == 8
assert loaded["rollups"]["controlled_apply_queue_count"] == 5
assert loaded["rollups"]["critical_break_glass_queue_count"] == 2
assert loaded["rollups"]["owner_response_required_count"] == 2
assert loaded["rollups"]["high_risk_owner_review_required_count"] == 0
assert loaded["rollups"]["telegram_send_count"] == 0
@@ -118,6 +130,26 @@ def test_ai_agent_high_risk_owner_review_queue_requires_rollback_owner(tmp_path)
load_latest_ai_agent_high_risk_owner_review_queue(tmp_path)
def test_ai_agent_high_risk_owner_review_queue_keeps_high_risk_on_controlled_apply(tmp_path):
snapshot = _snapshot()
high_item = next(item for item in snapshot["owner_review_queue_items"] if item["risk_tier"] == "high")
high_item["owner_response_required"] = True
_write_snapshot(tmp_path, snapshot)
with pytest.raises(ValueError, match="owner_response_required"):
load_latest_ai_agent_high_risk_owner_review_queue(tmp_path)
def test_ai_agent_high_risk_owner_review_queue_keeps_critical_on_break_glass(tmp_path):
snapshot = _snapshot()
critical_item = next(item for item in snapshot["owner_review_queue_items"] if item["risk_tier"] == "critical")
critical_item["owner_response_required"] = False
_write_snapshot(tmp_path, snapshot)
with pytest.raises(ValueError, match="owner_response_required"):
load_latest_ai_agent_high_risk_owner_review_queue(tmp_path)
def test_ai_agent_high_risk_owner_review_queue_requires_rollup_consistency(tmp_path):
snapshot = _snapshot()
snapshot["rollups"]["queue_item_count"] = 99

25
apps/api/tests/test_ai_agent_high_risk_owner_review_queue_api.py
View File

@@ -21,8 +21,12 @@ def test_ai_agent_high_risk_owner_review_queue_endpoint_returns_committed_snapsh
assert data["program_status"]["read_only_mode"] is True
assert (
data["program_status"]["runtime_authority"]
== "high_risk_owner_review_queue_no_live_execution_committed_snapshot"
== "controlled_apply_break_glass_queue_readback_no_live_execution"
)
assert data["queue_truth"]["all_high_risk_actions_paused"] is False
assert data["queue_truth"]["high_risk_owner_review_required"] is False
assert data["queue_truth"]["high_risk_controlled_apply_enabled"] is True
assert data["queue_truth"]["critical_break_glass_required"] is True
assert data["rollups"]["source_readback_count"] == len(data["source_readbacks"]) == 6
assert data["rollups"]["queue_item_count"] == len(data["owner_review_queue_items"]) == 7
assert data["rollups"]["high_risk_queue_count"] == 5
@@ -34,6 +38,10 @@ def test_ai_agent_high_risk_owner_review_queue_endpoint_returns_committed_snapsh
assert data["rollups"]["rollback_owner_required_count"] == 7
assert data["rollups"]["postcheck_required_count"] == 7
assert data["rollups"]["blocked_runtime_action_count"] == 42
assert data["rollups"]["controlled_apply_queue_count"] == 5
assert data["rollups"]["critical_break_glass_queue_count"] == 2
assert data["rollups"]["owner_response_required_count"] == 2
assert data["rollups"]["high_risk_owner_review_required_count"] == 0
assert data["rollups"]["owner_response_received_count"] == 0
assert data["rollups"]["owner_response_accepted_count"] == 0
assert data["rollups"]["auto_worker_run_count"] == 0
@@ -52,8 +60,19 @@ def test_ai_agent_high_risk_owner_review_queue_endpoint_returns_committed_snapsh
assert data["telegram_policy"]["telegram_send_allowed"] is False
assert data["telegram_policy"]["gateway_queue_write_allowed"] is False
assert data["telegram_policy"]["direct_bot_api_allowed"] is False
assert data["routing_policy"]["high_risk_default_route"] == "pause_to_owner_review_queue"
assert data["routing_policy"]["critical_risk_default_route"] == "pause_to_owner_review_queue"
assert data["routing_policy"]["high_risk_default_route"] == "controlled_apply_queue"
assert data["routing_policy"]["critical_risk_default_route"] == "critical_break_glass_queue"
assert data["routing_policy"]["owner_response_required"] is False
assert data["routing_policy"]["verbal_approval_accepted"] is False
assert all(
item["owner_response_required"] is False
for item in data["owner_review_queue_items"]
if item["risk_tier"] == "high"
)
assert all(
item["owner_response_required"] is True
for item in data["owner_review_queue_items"]
if item["risk_tier"] == "critical"
)
assert data["activation_boundaries"]["live_execution_enabled"] is False
assert data["activation_boundaries"]["openclaw_replacement_allowed"] is False