Files
mini-openclaw-v2/backend/test_card_to_run.py

217 lines
11 KiB
Python

"""纠偏第二批 B-c · 卡片选入方案并影响核查(BR-49 / AC-52)。
审计(audit.json P1-3)确认:知识卡片此前没有任何调用方,"卡片影响核查"只存在于
孤立的单元测试里。本测试走**真实执行链**:
提炼/人工卡片 → 隔离评测 → 独立审核发布 → 装配方案 → 发布方案
→ service.enqueue → DurableWorker.process_next() → 运行结果
用**同项目两次运行对照**证明卡片确实改变了核查结果:基座方案里没有该卡片,
新方案里有;两次运行都能读出各自的 `component_versions` 与核查项依据。
不用注入记录;临时库;不触碰真实数据根。
"""
import os
import tempfile
import unittest
from pathlib import Path
from unittest.mock import patch
from fastapi import FastAPI
from fastapi.testclient import TestClient
from api import prescreen as prescreen_api
from api import project_assistant as api
from api import risk_agent_config as catalog
from services import assistant_store as store
from services import knowledge_cards
from services.assistant_worker import DurableWorker
from services.project_assistant import ProjectAssistantService
PROJECT = "DEMO-B"
BASE = "/api/xingbang/risk-agent/project-assistant"
PRESCREEN = BASE + "/prescreen"
MANAGER = {"X-Xingbang-Demo-Role": "manager"}
ADMIN = {"X-Xingbang-Demo-Role": "admin"}
PUBLISHER = {"Authorization": "Bearer fixture-publisher"}
CARD_ID = "skill.history.repayment"
CARD_VERSION = "1.0.0"
CARD_CODE = "history.repayment.dependency"
# 自定义 Skill 的发布门禁要求独立测试样本(component_evaluation 的既有规则)
TEST_CASES = [
{"name": "缺证样本", "values": {"financial.operating_cashflow": None},
"expected_outputs": {}, "expected_checks": {CARD_CODE: "unknown"}},
{"name": "已提供样本", "values": {"financial.operating_cashflow": "1200"},
"expected_outputs": {}, "expected_checks": {CARD_CODE: "pass"}},
]
class CardToRunTests(unittest.TestCase):
def setUp(self):
self.temp = tempfile.TemporaryDirectory(ignore_cleanup_errors=True)
self.path = Path(self.temp.name)
self.service = ProjectAssistantService(self.path / "assistant.sqlite3")
self.patches = [
patch.object(api, "service", lambda: self.service),
patch.object(prescreen_api, "_service", lambda: self.service),
patch.object(catalog, "DB_PATH", self.path / "catalog.sqlite3"),
patch.dict(os.environ, {"XINGBANG_AUTH_USERS": '[{"id":"config-admin-b","role":"admin","projects":["DEMO-B"],"token":"fixture-publisher"}]', "XINGBANG_REQUIRE_AUTH": "0",
"XINGBANG_LOCAL_MODEL_URL": "", "XINGBANG_DEMO_ROLE_LOGIN": "1"}),
]
for item in self.patches:
item.start()
self.worker = DurableWorker(self.service)
app = FastAPI()
app.include_router(api.router)
app.include_router(prescreen_api.router)
self.client = TestClient(app)
def tearDown(self):
self.worker.stop()
self.client.close()
for item in reversed(self.patches):
item.stop()
self.temp.cleanup()
# ------------------------------------------------------------ 工具
def call(self, method, path, body=None, headers=MANAGER, root=PRESCREEN):
return self.client.request(method, root + path, json=body, headers=headers)
def propose_card(self):
with self.service.connect() as db:
return knowledge_cards.propose(db, {
"id": CARD_ID, "kind": "skill", "name": "核查回款集中依赖",
"engine": "required", "inputs": ["financial.operating_cashflow"],
"code": CARD_CODE, "title": "核查回款集中依赖",
"test_cases": TEST_CASES,
}, "demo-manager", PROJECT, origin="history_derived", source_event_ids=["event-1"])
def publish(self, path, headers=PUBLISHER):
evaluated = self.call("POST", path, {"action": "evaluate"}, headers)
self.assertEqual(evaluated.status_code, 200, evaluated.text)
return self.call("POST", path, {"action": "publish"}, headers)
def assemble_plan(self, skills, name="含历史经验卡的预审方案"):
return self.call("POST", f"/projects/{PROJECT}/review-plans", {
"name": name, "skills": skills, "rules": ["demo.coverage"],
"template": "template.ai-pre-review",
"budget": {"timeout_seconds": 180, "max_tool_calls": 18},
"audience": ["manager"], "run_kind": "pre_review",
}, headers=ADMIN)
def run_with(self, plan_id):
queued = self.call("POST", f"/projects/{PROJECT}/runs",
{"requirement": "核对回款与现金流", "plan_id": plan_id},
root=BASE)
self.assertIn(queued.status_code, (200, 202), queued.text)
run_id = queued.json()["id"]
self.assertTrue(self.worker.process_next(), "队列里应当有一次待执行运行")
with self.service.connect() as db:
return store.get(db, "run", run_id)
@staticmethod
def checks_of(run, code):
return [check for step in run["steps"] for check in step.get("findings", [])
if check.get("code") == code]
# ------------------------------------------------------------ AC-52
def test_published_card_changes_the_run_and_is_traceable(self):
# 1) 历史提炼候选 → 评测 → 独立审核发布
card = self.propose_card()
published = self.publish(f"/knowledge-candidates/{CARD_ID}/transitions")
self.assertEqual(published.status_code, 200, published.text)
self.assertEqual(published.json()["version"], CARD_VERSION)
# 2) 基座方案:卡片不在方案里 → 运行里不应出现该核查项
baseline = self.run_with("corporate.pre-review")
self.assertEqual(self.checks_of(baseline, CARD_CODE), [],
"卡片没进方案时不得凭空产生该核查项")
# 3) 把已发布卡片选入方案并发布
plan = self.assemble_plan(
["corporate.consistency", "corporate.cashflow", "corporate.assets",
"corporate.coverage", CARD_ID])
self.assertEqual(plan.status_code, 201, plan.text)
plan_id = plan.json()["component_id"]
self.assertEqual(self.publish(f"/review-plans/{plan_id}/transitions").status_code, 200)
# 4) 用新方案真实运行
run = self.run_with(plan_id)
refs = {(item["id"], item["version"]) for item in run["component_versions"]}
self.assertIn((CARD_ID, CARD_VERSION), refs, "运行必须冻结并记录卡片版本")
checks = self.checks_of(run, CARD_CODE)
self.assertEqual(len(checks), 1, "已发布卡片必须真的产生核查项")
step = next(step for step in run["steps"] if step.get("skill", {}).get("id") == CARD_ID)
self.assertEqual(step["status"] in ("completed", "partial"), True)
self.assertEqual(step["findings"][0]["basis_ids"], [f"{CARD_ID}@{CARD_VERSION}"],
"核查项必须能追溯到卡片版本")
self.assertEqual(step["findings"][0]["fields"], ["financial.operating_cashflow"])
# 5) 旧运行仍可读,且没有被新方案改写
self.assertEqual(self.checks_of(baseline, CARD_CODE), [])
with self.service.connect() as db:
self.assertNotEqual(store.get(db, "run", baseline["id"])["status"], "")
def test_draft_card_cannot_be_selected_into_a_plan(self):
"""草稿卡片不得进入新运行(BR-27 / AC-28)——装配阶段就要拒。"""
self.propose_card() # 只提议,不发布
plan = self.assemble_plan(["corporate.coverage", CARD_ID])
self.assertEqual(plan.status_code, 201, plan.text)
# 发布时依赖解析拒绝未发布组件
evaluated = self.call("POST", f"/review-plans/{plan.json()['component_id']}/transitions",
{"action": "evaluate"}, PUBLISHER)
self.assertEqual(evaluated.status_code, 200, evaluated.text)
published = self.call("POST", f"/review-plans/{plan.json()['component_id']}/transitions",
{"action": "publish"}, PUBLISHER)
self.assertEqual(published.status_code, 409, published.text)
def test_missing_dependency_fails_evaluation_instead_of_being_dropped(self):
"""方案引用不存在的组件 → 评测不通过、发布被拒,不静默丢掉该项。"""
self.propose_card()
plan = self.assemble_plan(["corporate.coverage", "skill.not.exists"])
self.assertEqual(plan.status_code, 201, plan.text)
plan_ref = plan.json()["component_id"]
evaluated = self.call("POST", f"/review-plans/{plan_ref}/transitions",
{"action": "evaluate"}, PUBLISHER)
self.assertEqual(evaluated.status_code, 200, evaluated.text)
self.assertFalse(evaluated.json()["passed"], "依赖缺失必须让评测不通过")
published = self.call("POST", f"/review-plans/{plan_ref}/transitions",
{"action": "publish"}, PUBLISHER)
self.assertEqual(published.status_code, 409, published.text)
def test_plan_rejects_tool_configuration_surface(self):
"""ADR-11:方案不得自带工具配置面,业务侧不能借方案扩权。"""
response = self.call("POST", f"/projects/{PROJECT}/review-plans", {
"name": "越权方案", "skills": ["corporate.coverage"],
"template": "template.ai-pre-review", "allowed_tools": ["execute_sql"],
})
self.assertEqual(response.status_code, 422)
self.assertIn("allowed_tools", response.text)
def test_transitioning_an_unknown_plan_is_404(self):
"""不存在的方案必须是 404,不是 500(归属查询在 run() 内完成)。"""
response = self.call("POST", "/review-plans/plan.not.exists/transitions",
{"action": "evaluate"}, PUBLISHER)
self.assertEqual(response.status_code, 404)
self.assertEqual(response.json()["error"]["code"], "NOT_FOUND")
def test_listed_plans_show_the_selected_cards(self):
self.propose_card()
self.publish(f"/knowledge-candidates/{CARD_ID}/transitions")
plan = self.assemble_plan(["corporate.coverage", CARD_ID])
listed = self.call("GET", f"/projects/{PROJECT}/review-plans").json()["items"]
mine = next(item for item in listed if item["id"] == plan.json()["component_id"])
self.assertIn(CARD_ID, mine["skills"], "方案列表必须显示选入的卡片")
self.assertEqual(mine["status"], "draft")
if __name__ == "__main__":
unittest.main()