mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 02:01:57 +00:00
ThearchyHelios
941723b17c
- 身份与工具：登录页建第一个管理员（认人接口的 401 说明能不能建）；搭建者有只读的人和组名录，选人给候选并核对名字；授权三行带标题；端点请求头名校验（空名、非法名存不进、发不出时说清原因）；写入能力在工具页试运行一律演练、不发送，HTTP 状态照实说成败；写记忆不算只读、查询能力只能选同一端点上的只读能力；审批与遮罩的说法和实际规则一致；开登录时不要署名、导入和发布记登录的人；使用者不显示推演等只给搭建者的操作，403 按是否开登录说话；登录页只认一次人 - 流程与实例：图里用到的岗位（派给谁、超时升级给谁、发送确认派给谁）全部派生、没绑人的发布时报错；写外部系统的参数按能力声明的类型收拢后再算哈希和发送，收不成就不发；发起人的实例页只看这一步的状态，处理人才给待办卡；会签进度带登录核实和时间，升级后照实写规则；实例状态写流程的结论（已退回、未确认、等待核实写入结果），核实生效后不再挂失败原因；流程应用首页不派生最近封存运行的结果块，看板没数据时是安静的空态；样例⑩的金额带类型和下限、待办标题带金额，表单上下限前后端一起生效 - 看板与证据：定时任务模板加 ISO 周、上一 ISO 周、本月、上月；出处链穿过原样解析查询结果的整形节点，追不到原始数据的照实说停在哪里，样例⑨改用 cell()；应用首页和版本历史都能做成应用；指标块单位不重复；版本列表带退役说明；定时任务页头跟着刷新；删视图要确认、新视图能起名；编辑器属性用中文名、术语带解释 - 指标出具（用户 2026-10-15 拍板）：契约只写「指标来自」、上游没有报告撰写节点、成果里没有模型写的话时按指标出具——必需指标齐、指标输入追到原始数据、成果字段里的数回指得上才完整出具；受管级要求每个指标静态地追到原始数据（governed.metric_untraced）；出具横幅和应用印章写明按指标 - 受管工作流降为已发布只给管理员（Q8 顺延，服务端同样拦）；工作流接口带当前发布版的级别，画布上改过之后也认得出受管工作流
252 lines
13 KiB
Python
252 lines
13 KiB
Python
"""口径卡的输入追没追到原始数据（engine/lineage.py）：一处认定，运行时、证据面板、应用检查三处用。
|
|
|
|
- 「有出处」说的是数由口径卡确定性算出、复算一致；「追到原始数据」说的是输入一路对到查询快照里的格（或运行输入）。
|
|
证据接口的指标步骤另给 untraced：追不到的输入停在哪个节点（名字、类型），面板据此照实说，不能只写「有出处」
|
|
- 数据整形只是原样解析调用工具节点的结果（模板整串就是 {{ nodes.fetch }}）时，vars.q.rows[i][j] 穿过它接上快照；
|
|
表达式整形、模板带过滤器、包了一层的都不算
|
|
- 受管级应用里绑定的指标，输入追不到原始数据的，应用检查给提示（app.metric_untraced，警告）；已发布级不提示
|
|
|
|
库和数字都是编的。
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import sqlite3
|
|
import uuid
|
|
from types import SimpleNamespace
|
|
|
|
import pytest
|
|
from httpx import ASGITransport, AsyncClient
|
|
|
|
from app.apps.lint import _Issues, _lint_components
|
|
from app.apps.tree import node_map
|
|
from app.data.engine import engines
|
|
from app.db.base import SessionLocal
|
|
from app.db.models import DataSource, Run
|
|
from app.engine.lineage import RowCell, Stop, parsed_tool, row_cell, static_stops, traced, untraced
|
|
from app.engine.runner import run_manager
|
|
from app.main import app
|
|
|
|
SQL = "SELECT SUM(amount) AS gmv, COUNT(*) AS orders FROM orders"
|
|
|
|
|
|
def node(nid, ntype, label=None, **config):
|
|
return {"id": nid, "type": ntype, "position": {"x": 0, "y": 0}, "data": {"label": label or nid, "config": config}}
|
|
|
|
|
|
def chain(*nodes):
|
|
return {"nodes": list(nodes), "edges": [{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]}
|
|
|
|
|
|
# --------------------------------------------------------------------------
|
|
# 纯函数
|
|
# --------------------------------------------------------------------------
|
|
|
|
|
|
@pytest.mark.parametrize("path, want", [
|
|
("vars.q.rows[0][0]", RowCell("vars", "q", 0, 0)),
|
|
("nodes.parse.rows[2]['gmv']", RowCell("nodes", "parse", 2, "gmv")),
|
|
("vars.q.rows[-1][0]", None), # 负数下标：数的是取回的最后一行，对不到确定的行号
|
|
("vars.q.rows[vars.i][0]", None), # 下标是变量
|
|
("vars.q.rows[0]", None), # 整行
|
|
("vars.w.res.rows[0][0]", None), # 多包了一层
|
|
("cell(nodes.fetch, 0, 'gmv')", None),
|
|
("input.week", None),
|
|
])
|
|
def test_row_cell(path, want):
|
|
assert row_cell(path) == want
|
|
|
|
|
|
@pytest.mark.parametrize("config, want", [
|
|
({"mode": "json", "template": "{{ nodes.fetch }}"}, "fetch"),
|
|
({"mode": "json", "template": " {{nodes.fetch}}\n"}, "fetch"),
|
|
({"mode": "json", "template": "{{ nodes.fetch | trim }}"}, None), # 带过滤器：不算原样
|
|
({"mode": "json", "template": '{"res": {{ nodes.fetch }}}'}, None), # 包了一层
|
|
({"mode": "json", "template": "{{ nodes.note }}"}, None), # 不是调用工具节点
|
|
({"mode": "template", "template": "{{ nodes.fetch }}"}, None),
|
|
({"mode": "expression", "expression": "nodes.fetch"}, None),
|
|
])
|
|
def test_parsed_tool(config, want):
|
|
types = {"fetch": "tool", "note": "llm"}
|
|
assert parsed_tool("transform", config, types.get) == want
|
|
assert parsed_tool("code", config, types.get) is None
|
|
|
|
|
|
def test_traced_and_untraced():
|
|
cell = {"path": "cell(nodes.fetch, 0, 'gmv')", "value": 1, "via": "tool_cell", "artifact": "a", "status": "verified"}
|
|
mismatch = {**cell, "status": "mismatch"} # 对得到快照，只是值不一样：另有说法，出处是追到了的
|
|
lost = {**cell, "status": "unresolved", "artifact": None} # 找不到快照
|
|
field = {"path": "vars.kpi.gmv", "value": 2, "via": "agent_field", "artifact": "a", "status": "verified"}
|
|
run_input = {"path": "input.target", "value": 3, "via": "input", "status": "ok"}
|
|
shaped = {"path": "vars.t.x", "value": 4, "node_id": "mid", "via": "transform", "status": "ok"}
|
|
empty = {"path": "vars.t.y", "value": None, "node_id": "mid", "via": "transform", "status": "missing"}
|
|
assert [traced(i) for i in (cell, mismatch, lost, field, run_input, shaped)] == [True, True, False, True, True, False]
|
|
# 没有值的不算在追不到里：缺输入另有说法
|
|
assert untraced([cell, lost, field, run_input, shaped, empty, "坏数据"]) == [lost, shaped]
|
|
|
|
|
|
def graph_nodes():
|
|
return node_map(chain(
|
|
node("start", "input", "开始", fields=[{"name": "target"}]),
|
|
node("fetch", "tool", "取数", tool="db_query__shop", args={"sql": SQL}),
|
|
node("parse", "transform", "解析", mode="json", template="{{ nodes.fetch }}", assign_to="q"),
|
|
node("raw_n", "transform", "解析（过滤器）", mode="json", template="{{ nodes.fetch | trim }}", assign_to="raw"),
|
|
node("calc", "transform", "加工", mode="expression", expression="{'x': 1}", assign_to="t"),
|
|
node("bot", "agent", "取数助手", output_schema={"type": "object"}, cite_fields=True, assign_to="kpi"),
|
|
node("loose", "agent", "没核对的助手", output_schema={"type": "object"}, assign_to="guess"),
|
|
node("box", "code", "沙箱", evidence_role="source", assign_to="c"),
|
|
node("card", "metrics", "口径卡", metrics=[]),
|
|
))
|
|
|
|
|
|
@pytest.mark.parametrize("expression, stops", [
|
|
("cell(nodes.fetch, 0, 'gmv')", []),
|
|
("cell(vars.q, 0, 'gmv') + vars.q.rows[0][1]", []), # 原样解析的结果：按格、按位置都追到快照
|
|
("nodes.parse.rows[0]['gmv']", []),
|
|
("input.target + vars.target", []), # 运行输入
|
|
("vars.kpi.gmv + nodes.bot.data.gmv", []), # 开了按出处核对字段的 Agent
|
|
("vars.raw.rows[0][0]", [Stop("vars.raw.rows[0][0]", "raw_n")]),
|
|
("cell(vars.t, 0, 0)", [Stop("cell(vars.t, 0, 0)", "calc")]),
|
|
("vars.t.x * 2", [Stop("vars.t.x", "calc")]),
|
|
("vars.guess.gmv", [Stop("vars.guess.gmv", "loose")]),
|
|
("vars.c.gmv", [Stop("vars.c.gmv", "box")]), # 沙箱取数核对不到快照
|
|
("vars.nowhere + 1", [Stop("vars.nowhere", None)]),
|
|
("vars.t.x +", []), # 写错了的表达式由画布的校验报，这里不报
|
|
])
|
|
def test_static_stops(expression, stops):
|
|
assert static_stops(graph_nodes(), expression) == stops
|
|
|
|
|
|
# --------------------------------------------------------------------------
|
|
# 证据接口：指标步骤说追没追到原始数据
|
|
# --------------------------------------------------------------------------
|
|
|
|
|
|
@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
|
|
async def shop(tmp_path):
|
|
path = tmp_path / "shop.db"
|
|
with sqlite3.connect(path) as db:
|
|
db.executescript("CREATE TABLE orders (id INTEGER PRIMARY KEY, amount REAL);")
|
|
db.executemany("INSERT INTO orders VALUES (?, ?)", [(1, 100.5), (2, 200.0), (3, 300.0)])
|
|
name = f"origin_{uuid.uuid4().hex[:6]}"
|
|
async with SessionLocal() as session:
|
|
row = DataSource(name=name, kind="sqlite", database=str(path), readonly=True, enabled=True)
|
|
session.add(row)
|
|
await session.commit()
|
|
source_id = row.id
|
|
yield name
|
|
await engines.invalidate(source_id)
|
|
|
|
|
|
async def finish(graph: dict) -> Run:
|
|
run = await run_manager.start(graph=graph, input_payload={"target": 5})
|
|
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"运行 {run.id} 没跑完：{row.status}")
|
|
|
|
|
|
async def test_the_metric_step_says_how_far_the_inputs_were_traced(client, shop):
|
|
graph = chain(
|
|
node("start", "input", "开始", fields=[{"name": "target"}]),
|
|
node("fetch", "tool", "取数", tool=f"db_query__{shop}", args={"sql": SQL}),
|
|
node("parse", "transform", "解析查询结果", mode="json", template="{{ nodes.fetch }}", assign_to="q"),
|
|
node("raw_n", "transform", "整理查询结果", mode="json", template="{{ nodes.fetch | trim }}", assign_to="raw"),
|
|
node("card", "metrics", "口径卡", caliber="口径", metrics=[
|
|
{"id": "gmv", "name": "销售额", "decimals": 2, "expression": "vars.q.rows[0][0]"},
|
|
{"id": "raw", "name": "销售额（过滤器）", "decimals": 2, "expression": "vars.raw.rows[0][0] + input.target"},
|
|
{"id": "one", "name": "常量", "expression": "1"},
|
|
]),
|
|
node("out", "output", "成果", fields=[{"name": "r", "value": "{{ nodes.card.text }}"}]),
|
|
)
|
|
row = await finish(graph)
|
|
assert row.status == "succeeded", row.error
|
|
|
|
gmv = (await client.get(f"/api/runs/{row.id}/evidence/metrics/card/gmv")).json()["chain"]
|
|
step = gmv[0]
|
|
assert step["step"] == "metric" and step["recompute_ok"] is True and step["untraced"] == []
|
|
[inp] = [s for s in gmv if s["step"] == "input"]
|
|
assert (inp["via"], inp["through"], inp["through_label"], inp["status"]) == ("tool_cell", "parse", "解析查询结果",
|
|
"verified")
|
|
# 穿过整形节点接上了查询快照：链的最后一步是那次查询，高亮那一格
|
|
assert gmv[-1]["step"] == "query" and gmv[-1]["highlight"]["cells"] == [[0, "gmv"]]
|
|
|
|
raw = (await client.get(f"/api/runs/{row.id}/evidence/metrics/card/raw")).json()["chain"]
|
|
# 复算照样一致（有出处），可输入停在了整形节点：运行输入那一项追到了，不列
|
|
assert raw[0]["recompute_ok"] is True
|
|
assert raw[0]["untraced"] == [{"path": "vars.raw.rows[0][0]", "node_id": "raw_n", "label": "整理查询结果",
|
|
"type": "transform"}]
|
|
assert not [s for s in raw if s["step"] == "query"]
|
|
|
|
# 没有输入的常量：没有可追的，不给这个键
|
|
one = (await client.get(f"/api/runs/{row.id}/evidence/metrics/card/one")).json()["chain"]
|
|
assert "untraced" not in one[0]
|
|
|
|
|
|
# --------------------------------------------------------------------------
|
|
# 应用检查：受管级应用里绑定的指标
|
|
# --------------------------------------------------------------------------
|
|
|
|
|
|
def app_spec() -> dict:
|
|
return {
|
|
"format": "agentlab.app/1", "name": "看板", "pins": [{"alias": "main", "workflow_id": "wf", "version": 1,
|
|
"graph_hash": "h", "level": "governed", "use": "view"}],
|
|
"views": {"latest": {"source": {"kind": "latest_sealed", "pin": "main", "from": "formal"}}},
|
|
"pages": [{"id": "home", "title": "首页", "body": [
|
|
{"id": "m_gmv", "type": "metric", "props": {"label": "销售额"}, "bind": {"view": "latest", "metric": "card.gmv"}},
|
|
{"id": "m_raw", "type": "metric", "props": {"label": "整理过的"}, "bind": {"view": "latest", "metric": "card.raw"}},
|
|
{"id": "s_raw", "type": "series", "props": {"label": "趋势"}, "bind": {"view": "latest", "series": "card.raw"}},
|
|
]}],
|
|
}
|
|
|
|
|
|
def pinned_graph() -> dict:
|
|
return chain(
|
|
node("fetch", "tool", "取数", tool="db_query__shop", args={"sql": SQL}),
|
|
node("parse", "transform", "解析查询结果", mode="json", template="{{ nodes.fetch }}", assign_to="q"),
|
|
node("calc", "transform", "加工", mode="expression", expression="{'gmv': 1}", assign_to="t"),
|
|
node("card", "metrics", "口径卡", metrics=[
|
|
{"id": "gmv", "name": "销售额", "expression": "vars.q.rows[0][0] + cell(nodes.fetch, 0, 'orders')"},
|
|
{"id": "raw", "name": "加工后的销售额", "expression": "vars.t.gmv * 2"},
|
|
]),
|
|
)
|
|
|
|
|
|
def lint(level: str) -> list[dict]:
|
|
spec = app_spec()
|
|
pins = {"main": SimpleNamespace(row=object(), graph=pinned_graph(), level=level, label="周报 v1")}
|
|
out = _Issues()
|
|
_lint_components(spec, pins, spec["views"], level, out)
|
|
return [i for i in out.items if i["code"] == "app.metric_untraced"]
|
|
|
|
|
|
def test_governed_apps_flag_metrics_whose_inputs_stop_halfway():
|
|
found = lint("governed")
|
|
assert [(i["path"], i["level"], i["node_id"]) for i in found] == [
|
|
("pages[0].body[1].bind", "warning", "card"), ("pages[0].body[2].bind", "warning", "card")]
|
|
message = found[0]["message"]
|
|
assert "「加工后的销售额」" in message and "vars.t.gmv 停在「加工」（数据整形）" in message and "cell()" in message
|
|
# 原样解析的按位置取格、cell() 取的格都追到了快照：销售额不提示
|
|
assert not [i for i in found if i["path"].startswith("pages[0].body[0]")]
|
|
|
|
|
|
def test_published_apps_are_not_flagged():
|
|
assert lint("published") == []
|