Files
group_fqcd_jr/tools/chat_console.py
T
张胜宇 5d0becb67d 客服 Agent 重构收口:五出口决策链 + 知识库档位隔离 + 前端入参边界(答辩演示版本)
一、客服 Agent 智能增强(正面回应"不智能、动不动就转人工")
- 决策链由 2 个出口扩到 5 个:E1 澄清 / E2 计算型 / E3 知识直返 / E4 证据约束生成 / E5 分级回退
- 转人工从"默认动作"降为最后一档 E5c,只保留 4 类白名单:
  P0 反诈 / P1 账户与个人数据 / P2 写操作与争议 / 用户明确要求人工
- 46 条金标实测(修复前 → 修复后):
  转人工率 43.5% → 10.9%;出口准确率 45.7% → 100%;事实正确率 69.6% → 100%
  禁忌违反 1 → 0;档位越权 / 无出处数字 / 误拒 四项零容忍全 0
- 安全不变量 INV-1~INV-5;零容忍规则未删,改的是挂载点
  (输出侧字面黑名单 → 检索层档位隔离 + 判定层合规词表 + 输出守护)

二、知识库:档位单点化与物理隔离
- 新增 app/core/knowledge_tier.py 作为档位规则唯一落点(G-03),
  knowledge_contracts.py 原定义块改为显式再导出(X as X,非副本)
- 档位过滤由 bool 默认值(fail-open)改为 tiers 必填集合(缺参即 TypeError)
- Milvus 侧四集合按 visibility 分区键物理隔离;双 schema 收敛为一套
- 新增 app/core/actor.py:访客三元组与匿名判定的唯一构造/判定点(G-01/G-01b)
- 新增 app/core/fund_fee_rules.py:费率计算纯函数

三、前端入参边界对齐(本轮 W11 新修,4 处"校验宽于存储")
- message 加 max_length=8000(与浮窗 widget.js 的 maxlength 一致)
- session_id 加 1—64;idempotency_key 上限 128 → 64(对齐列宽 String(64))
- feedback_type 加 max_length=32(对齐列宽 String(32))
- 8 条路径参数补 min_length=1 + max_length=64 + 字符集正则
  ({session_id} / {run_id} / {handover_id})
- 改前超限值会落到 MySQL 才失败(500);改后一律 422 AGENT_INPUT_INVALID + 字段级定位
- 新增 tests/unit/api/test_frontend_boundaries.py(33 例),含"端点表 ↔ OpenAPI 全量对照"

四、投顾模块整体清除(D4.4 / D4.5)
- 删除投顾相关 controller / schema / model / repository / service 及门户页面
- tools/portal_api_check.py 同步作废 AD003/AD005/AD011/A047 四条用例与 advisor_t 登录
  (端点与账号均已不存在,此前稳定报 3 条假红)

五、验证(提交前实测)
- pytest -q:1856 passed / 2 skipped / 0 failed
- ruff check app tools tests:19(= 基线);mypy app:2(= 基线)
- 前端接口契约体检 portal_api_check.py:38 项,通过 34,失败 0,跳过 4
- 全链路冒烟 e2e_smoke_test.py --read-only:31/31
- HTTP 全链路探针 http_probe.py:11/11 succeeded
- 跨文档一致性 _consistency.py:GATE PASS
- 真机边界复验 12 条:12/12 符合预期

六、纪律与文档
- 可改文件白名单 A-09(docs/46)与底座会签申请单 A-10(docs/47,组 1—组 4 全部受理)
- 零 DDL:未新增/修改任何表结构,89 张业务表与基线一致
- 证据留痕:docs/evidence/**(含 46 条金标 score、快照、清除与重建记录)
- 未提交(刻意排除,见提交说明):仓库内 客服agent/ 与 开发文档/ 是 2026-09-16 前的
  过期副本(Todolist 440 行 vs 权威 D2.1 1167 行),权威正本在仓库外;
  _chunks_report.txt 是 tools/build_knowledge_chunks.py 生成的本地产物
2026-09-20 14:33:30 +08:00

234 lines
9.3 KiB
Python

"""本地客服控制台:一个能直接聊天的最小调试前端。
**为什么是独立进程,而不是给底座加接口**:底座目前没有登录接口(按计划推迟)。任何
"让浏览器直接拿令牌"的做法——无论是一个 dev token 端点还是把私钥给前端——都等于把
"任意身份"开放给任何能访问服务的人。这里把令牌签发与调用全部留在**本进程内**
(私钥不出进程),前端只跟本控制台说话,底座代码一行不改、也没有新增任何后门路由。
跑法:
```powershell
D:\\conda\\envs\\jr_py313\\python.exe tools\\chat_console.py
# 浏览器打开 http://127.0.0.1:8098
```
**运行前请停掉常驻 Worker**:本控制台自己驱动这一条 run 执行(`WorkerRuntime().execute`),
而常驻 Worker 与它共享 `agent_run` 队列,会抢走任务,导致页面一直转圈。
"""
import argparse
import datetime as dt
import sys
import uuid
from pathlib import Path
from typing import Any
import httpx
import jwt
import uvicorn
from fastapi import FastAPI
from fastapi.responses import HTMLResponse
from app.core.config import get_settings
from app.main import create_app
from app.worker.runtime import WorkerRuntime
sys.stdout.reconfigure(errors="replace")
AGENT_TYPE = "customer_service"
# 兜底话术的固定开头。不能用"客服热线"这类词判断——正常的适当性回答里也会建议客户
# 拨打客服热线,那样会把"已经给出了结论"误报成"已引导人工"。
FALLBACK_MARK = "抱歉,这个问题我暂时无法给出准确答复"
PAGE = """<!DOCTYPE html>
<html lang="zh-CN">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>南方基金 · 智能客服(本地调试)</title>
<style>
* { box-sizing: border-box; }
body { margin: 0; font-family: -apple-system, "Segoe UI", "Microsoft YaHei", sans-serif;
background: #f2f4f7; display: flex; justify-content: center; height: 100vh; }
.wrap { width: min(760px, 100%); display: flex; flex-direction: column; height: 100vh; }
header { padding: 14px 18px; background: #1f2d3d; color: #fff; font-size: 15px; }
header small { opacity: .65; margin-left: 8px; font-size: 12px; }
#log { flex: 1; overflow-y: auto; padding: 18px; }
.row { display: flex; margin-bottom: 14px; }
.row.user { justify-content: flex-end; }
.bubble { max-width: 78%; padding: 11px 14px; border-radius: 10px; line-height: 1.6;
white-space: pre-wrap; word-break: break-word; font-size: 14px; }
.user .bubble { background: #2f6fed; color: #fff; border-bottom-right-radius: 3px; }
.bot .bubble { background: #fff; color: #1f2d3d; border-bottom-left-radius: 3px;
box-shadow: 0 1px 3px rgba(0,0,0,.08); }
.meta { font-size: 11px; color: #8894a4; margin: 5px 2px 0; }
.meta .tag { display: inline-block; padding: 1px 6px; border-radius: 4px;
background: #e6ebf2; margin-right: 6px; }
.meta .tag.warn { background: #fde8e8; color: #b42318; }
footer { padding: 12px 18px 18px; background: #fff; border-top: 1px solid #e4e9f0;
display: flex; gap: 10px; }
input { flex: 1; padding: 12px 14px; border: 1px solid #d5dbe4; border-radius: 8px;
font-size: 14px; outline: none; }
input:focus { border-color: #2f6fed; }
button { padding: 12px 22px; border: 0; border-radius: 8px; background: #2f6fed; color: #fff;
font-size: 14px; cursor: pointer; }
button:disabled { background: #a9bbe0; cursor: default; }
.hint { font-size: 12px; color: #8894a4; padding: 0 18px 8px; }
</style>
</head>
<body>
<div class="wrap">
<header>南方基金 · 智能客服 <small>本地调试控制台(回答全部来自公司资料库)</small></header>
<div id="log"></div>
<div class="hint">可以直接试:<em>基金赎回几天到账</em> · <em>季季盈90天起投多少</em> ·
<em>C1 客户能买什么</em> · 然后再问 <em>那它风险高吗</em>(试试多轮指代)</div>
<footer>
<input id="box" placeholder="说点什么…" autocomplete="off">
<button id="send">发送</button>
</footer>
</div>
<script>
const log = document.getElementById('log');
const box = document.getElementById('box');
const send = document.getElementById('send');
const sessionId = 'web-' + Math.random().toString(36).slice(2, 10);
let busy = false;
function add(cls, text, meta) {
const row = document.createElement('div');
row.className = 'row ' + cls;
const bubble = document.createElement('div');
bubble.className = 'bubble';
bubble.textContent = text;
const holder = document.createElement('div');
holder.appendChild(bubble);
if (meta) {
const m = document.createElement('div');
m.className = 'meta';
m.innerHTML = meta;
holder.appendChild(m);
}
row.appendChild(holder);
log.appendChild(row);
log.scrollTop = log.scrollHeight;
}
async function ask() {
const message = box.value.trim();
if (!message || busy) return;
busy = true; send.disabled = true;
add('user', message);
box.value = '';
add('bot', '正在查询公司资料…');
const pending = log.lastElementChild;
try {
const res = await fetch('/api/chat', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ message, session_id: sessionId }),
});
const data = await res.json();
pending.remove();
const tagIntent = '<span class="tag">意图 ' + (data.intent || '?') + '</span>';
const tagGuide = data.guided ? '<span class="tag warn">已引导人工</span>' : '';
add('bot', data.answer || '(空回答)', tagIntent + tagGuide);
} catch (err) {
pending.remove();
add('bot', '控制台出错:' + err, '');
} finally {
busy = false; send.disabled = false; box.focus();
}
}
send.onclick = ask;
box.addEventListener('keydown', (e) => { if (e.key === 'Enter') ask(); });
box.focus();
</script>
</body>
</html>
"""
def token(subject: str) -> str:
settings = get_settings()
private_key = Path(settings.jwt_private_key_path).read_text(encoding="utf-8")
now = dt.datetime.now(dt.UTC)
return jwt.encode(
{
"sub": subject, "iss": settings.jwt_issuer, "aud": settings.jwt_audience,
"exp": now + dt.timedelta(minutes=30), "nbf": now - dt.timedelta(seconds=5),
"jti": str(uuid.uuid4()),
},
private_key,
algorithm="RS256",
)
def build_console(user_id: str) -> FastAPI:
console = FastAPI(title="客服本地控制台")
# 底座以 in-process 方式挂载:不经过网络、不新增任何路由,私钥只在签发时用到
base_app = create_app()
auth = {"Authorization": f"Bearer {token(user_id)}"}
@console.get("/", response_class=HTMLResponse)
async def index() -> HTMLResponse:
return HTMLResponse(PAGE)
@console.post("/api/chat")
async def chat(payload: dict[str, Any]) -> dict[str, Any]:
message = str(payload.get("message") or "").strip()
session_id = str(payload.get("session_id") or f"web-{uuid.uuid4().hex[:8]}")
if not message:
return {"answer": "请先输入内容。", "intent": "", "guided": False}
async with httpx.AsyncClient(
transport=httpx.ASGITransport(app=base_app), base_url="http://base", timeout=120
) as client:
accepted = await client.post(
"/api/v1/agent-runs",
json={
"agent_type": AGENT_TYPE, "message": message,
"session_id": session_id, "idempotency_key": uuid.uuid4().hex,
},
headers=auth,
)
if accepted.status_code != 202:
body = accepted.text[:200]
return {"answer": f"受理失败({accepted.status_code}):{body}",
"intent": "", "guided": False}
run_id = accepted.json()["data"]["run_id"]
# 本控制台自己驱动这一条 run;若常驻 Worker 同时在跑会与它抢队列
await WorkerRuntime().execute(run_id)
detail = (await client.get(
f"/api/v1/agent-runs/{run_id}", headers=auth
)).json()["data"]
result = detail.get("result") or {}
answer = str(result.get("content") or "")
if detail.get("status") != "succeeded":
answer = answer or f"运行未成功:{detail.get('status')} {detail.get('error_code')}"
# intent 在结果里可能是 {"intent": ..., "confidence": ...},也可能已是字符串
raw_intent = result.get("intent")
intent = raw_intent.get("intent") if isinstance(raw_intent, dict) else raw_intent
return {
"answer": answer,
"intent": intent or "",
"guided": FALLBACK_MARK in answer,
}
return console
def main() -> None:
parser = argparse.ArgumentParser(description="客服本地控制台")
parser.add_argument("--port", type=int, default=8098)
parser.add_argument("--user", default="9001", help="以哪个客户身份对话(默认 9001)")
args = parser.parse_args()
print(f"控制台:http://127.0.0.1:{args.port} 身份={args.user}")
print("提示:运行前请停掉常驻 Worker,否则它会抢走任务、页面会一直转圈。")
uvicorn.run(build_console(args.user), host="127.0.0.1", port=args.port, log_level="warning")
if __name__ == "__main__":
main()