feat(governance): 新增 reviewer queue no-write readback
This commit is contained in:
110
apps/api/tests/test_ai_agent_reviewer_queue_no_write_readback.py
Normal file
110
apps/api/tests/test_ai_agent_reviewer_queue_no_write_readback.py
Normal file
@@ -0,0 +1,110 @@
|
||||
import copy
|
||||
import json
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
|
||||
from src.services.ai_agent_reviewer_queue_no_write_readback import (
|
||||
load_latest_ai_agent_reviewer_queue_no_write_readback,
|
||||
)
|
||||
|
||||
|
||||
REPO_ROOT = Path(__file__).resolve().parents[3]
|
||||
FIXTURE = REPO_ROOT / "docs/evaluations/ai_agent_reviewer_queue_no_write_readback_2026-06-13.json"
|
||||
|
||||
|
||||
def test_load_latest_ai_agent_reviewer_queue_no_write_readback_snapshot() -> None:
|
||||
data = load_latest_ai_agent_reviewer_queue_no_write_readback()
|
||||
|
||||
assert data["schema_version"] == "ai_agent_reviewer_queue_no_write_readback_v1"
|
||||
assert data["program_status"]["current_task_id"] == "P2-117"
|
||||
assert data["program_status"]["next_task_id"] == "P2-118"
|
||||
assert data["program_status"]["overall_completion_percent"] == 100
|
||||
assert data["program_status"]["read_only_mode"] is True
|
||||
|
||||
rollups = data["rollups"]
|
||||
assert rollups["reviewer_queue_readback_fixture_count"] == 5
|
||||
assert rollups["queue_item_mapping_count"] == 5
|
||||
assert rollups["readback_verifier_check_count"] == 5
|
||||
assert rollups["blocked_queue_write_count"] == 5
|
||||
assert rollups["operator_action_count"] == 5
|
||||
assert rollups["approval_required_fixture_count"] == 2
|
||||
assert rollups["blocked_fixture_count"] == 2
|
||||
assert rollups["approval_required_mapping_count"] == 1
|
||||
assert rollups["blocked_mapping_count"] == 2
|
||||
assert rollups["approval_required_verifier_count"] == 2
|
||||
assert rollups["critical_blocker_count"] == 3
|
||||
|
||||
assert {mapping["target_queue"] for mapping in data["queue_item_mappings"]} == {"reviewer_queue_preview"}
|
||||
assert {fixture["readback_mode"] for fixture in data["reviewer_queue_readback_fixtures"]} == {
|
||||
"no_write_fixture"
|
||||
}
|
||||
|
||||
zero_fields = [
|
||||
"owner_approval_received_count",
|
||||
"canonical_runtime_target_read_count",
|
||||
"live_query_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",
|
||||
]
|
||||
for field in zero_fields:
|
||||
assert rollups[field] == 0
|
||||
|
||||
|
||||
def test_reviewer_queue_no_write_readback_rejects_queue_write_enabled(tmp_path: Path) -> None:
|
||||
source = json.loads(FIXTURE.read_text(encoding="utf-8"))
|
||||
source["reviewer_queue_readback_fixtures"][0]["reviewer_queue_write_enabled"] = True
|
||||
target = tmp_path / "ai_agent_reviewer_queue_no_write_readback_2026-06-13.json"
|
||||
target.write_text(json.dumps(source), encoding="utf-8")
|
||||
|
||||
with pytest.raises(ValueError, match="must not enable reviewer queue write"):
|
||||
load_latest_ai_agent_reviewer_queue_no_write_readback(tmp_path)
|
||||
|
||||
|
||||
def test_reviewer_queue_no_write_readback_rejects_truth_write_flag(tmp_path: Path) -> None:
|
||||
source = json.loads(FIXTURE.read_text(encoding="utf-8"))
|
||||
source["readback_truth"]["reviewer_queue_write_enabled"] = True
|
||||
target = tmp_path / "ai_agent_reviewer_queue_no_write_readback_2026-06-13.json"
|
||||
target.write_text(json.dumps(source), encoding="utf-8")
|
||||
|
||||
with pytest.raises(ValueError, match="live read/send/write flags"):
|
||||
load_latest_ai_agent_reviewer_queue_no_write_readback(tmp_path)
|
||||
|
||||
|
||||
def test_reviewer_queue_no_write_readback_rejects_target_queue_drift(tmp_path: Path) -> None:
|
||||
source = json.loads(FIXTURE.read_text(encoding="utf-8"))
|
||||
source["queue_item_mappings"][0]["target_queue"] = "live_reviewer_queue"
|
||||
target = tmp_path / "ai_agent_reviewer_queue_no_write_readback_2026-06-13.json"
|
||||
target.write_text(json.dumps(source), encoding="utf-8")
|
||||
|
||||
with pytest.raises(ValueError, match="must target reviewer_queue_preview"):
|
||||
load_latest_ai_agent_reviewer_queue_no_write_readback(tmp_path)
|
||||
|
||||
|
||||
def test_reviewer_queue_no_write_readback_rejects_rollup_drift(tmp_path: Path) -> None:
|
||||
source = json.loads(FIXTURE.read_text(encoding="utf-8"))
|
||||
source["rollups"]["reviewer_queue_readback_fixture_count"] = 4
|
||||
target = tmp_path / "ai_agent_reviewer_queue_no_write_readback_2026-06-13.json"
|
||||
target.write_text(json.dumps(source), encoding="utf-8")
|
||||
|
||||
with pytest.raises(ValueError, match="rollup counts mismatch"):
|
||||
load_latest_ai_agent_reviewer_queue_no_write_readback(tmp_path)
|
||||
|
||||
|
||||
def test_reviewer_queue_no_write_readback_rejects_forbidden_display_terms(tmp_path: Path) -> None:
|
||||
source = copy.deepcopy(json.loads(FIXTURE.read_text(encoding="utf-8")))
|
||||
source["operator_actions"][0]["operator_instruction"] = "do not expose session_id"
|
||||
target = tmp_path / "ai_agent_reviewer_queue_no_write_readback_2026-06-13.json"
|
||||
target.write_text(json.dumps(source), encoding="utf-8")
|
||||
|
||||
with pytest.raises(ValueError, match="forbidden display terms"):
|
||||
load_latest_ai_agent_reviewer_queue_no_write_readback(tmp_path)
|
||||
@@ -0,0 +1,43 @@
|
||||
import pytest
|
||||
from httpx import ASGITransport, AsyncClient
|
||||
|
||||
from src.main import app
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_get_agent_reviewer_queue_no_write_readback_api() -> None:
|
||||
transport = ASGITransport(app=app)
|
||||
async with AsyncClient(transport=transport, base_url="http://test") as client:
|
||||
response = await client.get("/api/v1/agents/agent-reviewer-queue-no-write-readback")
|
||||
|
||||
assert response.status_code == 200
|
||||
data = response.json()
|
||||
assert data["schema_version"] == "ai_agent_reviewer_queue_no_write_readback_v1"
|
||||
assert data["program_status"]["current_task_id"] == "P2-117"
|
||||
assert data["program_status"]["next_task_id"] == "P2-118"
|
||||
assert data["program_status"]["overall_completion_percent"] == 100
|
||||
|
||||
rollups = data["rollups"]
|
||||
assert rollups["reviewer_queue_readback_fixture_count"] == 5
|
||||
assert rollups["queue_item_mapping_count"] == 5
|
||||
assert rollups["readback_verifier_check_count"] == 5
|
||||
assert rollups["blocked_queue_write_count"] == 5
|
||||
assert rollups["operator_action_count"] == 5
|
||||
assert rollups["critical_blocker_count"] == 3
|
||||
assert rollups["owner_approval_received_count"] == 0
|
||||
assert rollups["canonical_runtime_target_read_count"] == 0
|
||||
assert rollups["live_query_count"] == 0
|
||||
assert rollups["reviewer_queue_write_count"] == 0
|
||||
assert rollups["gateway_queue_write_count"] == 0
|
||||
assert rollups["telegram_send_count"] == 0
|
||||
assert rollups["bot_api_call_count"] == 0
|
||||
assert rollups["report_receipt_write_count"] == 0
|
||||
assert rollups["result_capture_write_count"] == 0
|
||||
assert rollups["learning_write_count"] == 0
|
||||
assert rollups["playbook_trust_write_count"] == 0
|
||||
assert rollups["production_write_count"] == 0
|
||||
|
||||
assert {mapping["target_queue"] for mapping in data["queue_item_mappings"]} == {"reviewer_queue_preview"}
|
||||
assert {fixture["reviewer_queue_write_enabled"] for fixture in data["reviewer_queue_readback_fixtures"]} == {
|
||||
False
|
||||
}
|
||||
Reference in New Issue
Block a user