Files
AgentLab/backend/tests/test_version_level.py
T
ThearchyHelios d071bf119f 可点击证据第三期:表名字段名和引文也点得开、可疑实体与结论句、受管门禁 G1–G5、证据审计导出
一、表结构、实体与引文

- 数据源工具查库时顺带存一份表结构快照(schema_snapshot),和查询快照一起进台账、进封存;
  台账里的查询记下 SQL 用到的表
- 报告里的表名、字段名可以点:反引号里的名字一律核对,正文里的裸名字只在明显是标识符时才自动链接,
  常见英文单词不链接;点开看到它出现在哪次查询、类型、所属的表,快照被截断时写明
- 找不到的名字标成可疑实体:在快照、SQL、结果列里都没有的记「可能是编造的名字」并给出最接近的 3 个;
  快照被截断、核对不了的记「核对不了」,不冤枉成编造
- 检索引文 [[q:K1|原文]]:原文在检索快照的命中片段里逐字出现(空白归一化后)才算有出处,点开高亮原文位置
- 自动链接的名字不算句子的引用,只有写作者写的引用才算

二、结论句与出具

- 报告节点新增 claims:off(默认,保持原样)或 require_citation(没有引用的结论句计入缺口)
- 受管级别的正式运行里,可疑实体和没有引用的结论句按「有缺口」降档,不拦截;探索运行和已发布级别只标注
- 契约的 claims 和报告节点的设置取更严的那个;受管正式运行里写 ignore 也按降档处理,并给出警告
- 出具横幅除了数字,也写出可疑实体和没有引用的结论句

三、受管门禁 G1–G5(只看图和配置)

- G1 带契约的出口,成果字段的文字只能来自报告撰写节点
- G2 模型写的文字不绕过报告撰写节点直接流进出口
- G3 报告撰写节点写明 numbers: strict、on_violation: fail、claims: require_citation
- G4 口径卡的输入不来自计算角色的沙箱代码
- G5 给口径卡供数的 agent 配 output_schema 并开 cite_fields
- 受管级别是错误、挡住发布;已发布级别只给警告;已经发布的版本照常运行,下次发布时按新规则检查
- 每条规则都接进发布前自动修复(一键修复 / 给选项 / 交给 Copilot);禁止的改动补上关 cite_fields、
  删 output_schema、放宽契约 claims;Copilot 兜底不许自己把代码节点标成 source、不许放宽报告节点
- 发布弹窗里选「交给 Copilot」时明确写「正在交给 Copilot」

四、旧运行与审计

- 没有契约的旧答案:数字和已封存的查询单元格、口径卡值按数值比对,每个数最多 3 个候选,
  默认折叠,写明「猜测的来源,不能当证据」
- 有契约的旧运行:候选只从已封存的口径卡里找,人为插进工件表的行不当证据
- 证据审计:GET /api/runs/{id}/evidence/audit 按「无证据 / 可疑实体 / 有出处 / 旧运行猜测」分组,
  可导出 JSON 和 CSV(带封存状态);记录页和画布上的报告节点显示「有出处 N · 无证据 M」

五、SQL 守卫修正(安全)

- 去注释改成单遍扫描,认得字符串和各方言的引号(单引号转义、MySQL 反斜杠与 #、PostgreSQL 的 $$ 与嵌套注释、
  Oracle 的 q'[…]'、SQLite 的方括号);以前字符串里的 -- 会被当成注释,SELECT '--'; DROP … 这类写法能骗过多语句拦截
- 看不到服务器设置时每种读法都扫一遍,读法之间切分不一致就拒绝,任何一种读法是写操作就按写操作处理
- 执行查询和工具危险判定都按数据源的方言检查

六、其他

- 修掉检查脚本里画布检查的一处竞态,并行估时按实测更新
2026-09-29 05:26:34 +08:00

160 lines
7.3 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""发布级别记在版本上(WorkflowVersion.level)。
以前「这次正式运行按不按受管出具」看的是工作流当前的 status。可 status 说的是当前画布:
受管版本发布之后在画布上改一笔,status 就退回 draft,再从那个受管版本发起的正式运行
就被当成已发布级别出具(单元格不必在契约里声明 cells 就放行)。级别要跟着版本走:
发布时记在那一版上,正式运行看它钉的版本;没有记过级别的老版本照旧看 status。
"""
from __future__ import annotations
import asyncio
import pytest
from httpx import ASGITransport, AsyncClient
from langchain_core.messages import AIMessage
from sqlalchemy import create_engine, select, text
from app.db.base import SessionLocal
from app.db.models import Run, RunEvent, Workflow, WorkflowVersion
from app.engine.runner import run_manager
from app.main import app
@pytest.fixture(autouse=True)
async def engine_up():
await run_manager.setup()
yield
await run_manager.shutdown()
@pytest.fixture
async def client():
async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as c:
yield c
@pytest.fixture(autouse=True)
def writer(monkeypatch):
from app.providers import mock_model
monkeypatch.setattr(mock_model.MockChatModel, "_decide",
lambda model, messages: AIMessage(content="本周销售额 [[m:gmv]]。"))
def node(nid, ntype, **config):
return {"id": nid, "type": ntype, "position": {"x": 0, "y": 0}, "data": {"label": nid, "config": config}}
def weekly() -> dict:
nodes = [
node("start", "input"),
node("fetch", "transform", mode="expression", expression="{'gmv': 300.5}", assign_to="kpi"),
node("card", "metrics", caliber="周报口径", metrics=[
{"id": "gmv", "name": "销售额", "unit": "元", "expression": "vars.kpi.gmv"}]),
node("write", "report", instructions="写周报",
numbers="strict", on_violation="fail", claims="require_citation"),
node("out", "output", fields=[{"name": "周报", "value": "{{ nodes.write.text }}"}],
contract={"report_from": "write", "metrics_from": ["card"], "required": ["gmv"], "strict": True}),
]
return {"nodes": nodes, "edges": [{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]}
async def version_row(wf: str, version: int) -> WorkflowVersion:
async with SessionLocal() as session:
return (await session.execute(select(WorkflowVersion).where(
WorkflowVersion.workflow_id == wf, WorkflowVersion.version == version))).scalar_one()
async def notes(run_id: str) -> list[dict]:
async with SessionLocal() as session:
rows = (await session.execute(select(RunEvent).where(
RunEvent.run_id == run_id, RunEvent.type == "log").order_by(RunEvent.seq))).scalars()
return [e.data for e in rows if e.data.get("code") == "run_governance"]
async def settle(run_id: str) -> Run:
for _ in range(400):
await asyncio.sleep(0.05)
async with SessionLocal() as session:
row = await session.get(Run, run_id)
if row.status in ("succeeded", "failed") and row.manifest_seq is not None:
return row
raise AssertionError(f"run {run_id} 没跑完:{row.status}")
@pytest.mark.parametrize("level", ["published", "governed"])
async def test_publishing_records_the_level_on_the_version(client, level):
wf = (await client.post("/api/workflows", json={"name": f"周报-{level}", "graph": weekly()})).json()["id"]
body = (await client.post(f"/api/workflows/{wf}/publish", json={"level": level})).json()
assert body["ok"] is True, body["issues"]
assert (await version_row(wf, 1)).level == level
versions = (await client.get(f"/api/workflows/{wf}/versions")).json()
assert versions[0]["level"] == level
detail = (await client.get(f"/api/workflows/{wf}/versions/1")).json()
assert detail["level"] == level and detail["published"] is True
async def test_a_blocked_publish_records_nothing(client):
graph = weekly()
graph["nodes"][-1]["data"]["config"]["contract"].pop("required")
wf = (await client.post("/api/workflows", json={"name": "周报-被拦", "graph": graph})).json()["id"]
assert (await client.post(f"/api/workflows/{wf}/publish", json={"level": "governed"})).json()["ok"] is False
assert (await version_row(wf, 1)).level is None
async def test_republishing_the_same_version_updates_its_level(client):
wf = (await client.post("/api/workflows", json={"name": "周报-升档", "graph": weekly()})).json()["id"]
await client.post(f"/api/workflows/{wf}/publish", json={"level": "published"})
await client.post(f"/api/workflows/{wf}/publish", json={"level": "governed"})
assert (await version_row(wf, 1)).level == "governed"
@pytest.mark.parametrize("level, governed", [("governed", True), ("published", False)])
async def test_editing_the_canvas_after_publishing_does_not_change_how_the_version_issues(client, level, governed):
from app.engine.governance import governed_formal
wf = (await client.post("/api/workflows", json={"name": f"周报-改画布-{level}", "graph": weekly()})).json()["id"]
assert (await client.post(f"/api/workflows/{wf}/publish", json={"level": level})).json()["ok"] is True
edited = weekly()
edited["nodes"][3]["data"]["label"] = "改过的报告"
after = (await client.patch(f"/api/workflows/{wf}", json={"graph": edited})).json()
assert after["status"] == "draft" and after["published_version"] == 1
r = await client.post("/api/runs", json={"workflow_id": wf, "run_class": "formal"})
assert r.status_code == 201, r.text
run_id = r.json()["id"]
row = await settle(run_id)
[note] = await notes(run_id)
assert note["governed"] is governed and "v1" in note["message"], note
assert await governed_formal(run_id) is governed
assert row.status == "succeeded", row.error
async def test_a_version_published_before_the_column_existed_follows_the_status(client):
"""老版本没有记级别:照旧看工作流现在的 status(test_contract_cells 那套旧规则)。"""
async with SessionLocal() as session:
w = Workflow(name="周报-老版本", graph=weekly(), status="governed", published_version=1)
session.add(w)
await session.flush()
session.add(WorkflowVersion(workflow_id=w.id, version=1, graph=weekly()))
await session.commit()
wf = w.id
r = await client.post("/api/runs", json={"workflow_id": wf, "run_class": "formal"})
run_id = r.json()["id"]
await settle(run_id)
[note] = await notes(run_id)
assert note["governed"] is True
def test_the_column_is_added_to_old_databases():
from app.db.base import _migrate
engine = create_engine("sqlite://")
with engine.begin() as conn:
conn.execute(text("CREATE TABLE workflow_versions (id VARCHAR(32) PRIMARY KEY, workflow_id VARCHAR(32), "
"version INTEGER, graph JSON, note TEXT, graph_hash VARCHAR(64))"))
conn.execute(text("INSERT INTO workflow_versions (id, workflow_id, version) VALUES ('a', 'w', 1)"))
_migrate(conn)
cols = {row[1] for row in conn.execute(text("PRAGMA table_info(workflow_versions)"))}
assert "level" in cols
assert conn.execute(text("SELECT level FROM workflow_versions")).scalar() is None