把求职者正在看的那条 JD，变成图谱里的一个新节点

// content.js —— 只做一件事：把页面上的字段取出来
// 站点适配器：每个站一套选择器；抽不到的字段返回 null，不猜
const ADAPTERS = {
  "www.liepin.com": () => ({
    title:      q(".job-title-box h1, h1"),
    company:    q(".company-name, .comp-name"),
    salary:     q(".job-salary, .salary"),
    tags:       qa(".labels span, .tag-list span"),
    duty:       block("岗位职责"),
    require:    block("任职要求")
  }),
  "www.zhipin.com": () => ({ /* 同上，选择器换成 BOSS 的 */ })
};

// 通用回退：站点改版时至少能拿到标题和薪资
function fallback() {
  return {
    title:   q("h1"),
    salary:  (document.body.innerText.match(/\\d+[-–]\\d+K/i) || [""])[0],
    duty:    block("岗位职责"),
    require: block("任职要求")
  };
}

// 监听后台的采集指令，并把结果回传
chrome.runtime.onMessage.addListener((msg, _sender, reply) => {
  if (msg.type !== "CAPTURE") return;
  const host = location.hostname.replace("www.", "");
  const pick = ADAPTERS[location.hostname] || fallback;   // 没适配就走回退
  const data = pick();
  reply({ ...data, url: location.href, capturedAt: Date.now(), site: host });
});

// background.js —— 打开侧栏、拿数据、提交服务端
const API = "https://your-domain.com";

// 1. 用户点工具栏图标：开侧栏 + 发起采集
chrome.action.onClicked.addListener(async (tab) => {
  await chrome.sidePanel.open({ tabId: tab.id });
  const data = await chrome.tabs.sendMessage(tab.id, { type: "CAPTURE" });
  await submit(data);                       // 2. 提交治理流水线
});

// 3. 提交：带用户 token，服务端校验身份后落原始池
async function submit(data) {
  const { token } = await chrome.storage.local.get("token");  // 复用 Web 端 JWT
  if (!token) return;                        // 没登录就不提交，不静默续期
  const res = await fetch(`${API}/api/jd/capture`, {
    method: "POST",
    headers: { "Content-Type": "application/json", Authorization: `Bearer ${token}` },
    body: JSON.stringify(data)
  });
  const { jd_id, confidence } = await res.json();   // 回给侧栏展示
  chrome.storage.session.set({ lastJd: { jd_id, confidence } });
}

# api.py —— 两个接口，复用已有的治理与诊断能力
@router.post("/api/jd/capture")
def capture(payload: JdIn, user=Depends(current_user)):
    # 1. 字段完整性校验：9 个必填项，缺关键字段直接拒绝
    if not payload.title or not payload.require:
        raise HTTPException(400, "字段不足，未进入流水线")

    # 2. SimHash 去重：命中已有文档就只更新 last_seen，不新增节点
    h = simhash(payload.text())
    if exist = store.find_by_simhash(h, threshold=3):
        exist.touch(source="extension")
        return { "jd_id": exist.id, "dedup": True }

    # 3. 和其他来源走同一条流水线，同一套阈值，不搞双标准
    jd = pipeline.ingest(payload, source="extension")

    # 4. 置信度不足的进人工复核队列，通过后才允许写图
    if jd.confidence < CONFIDENCE_GATE:
        review_queue.push(jd)
        return { "jd_id": jd.id, "confidence": jd.confidence, "need_review": True }

    graph.upsert_job(jd)                    # 5. 岗位节点 upsert，技能边 MERGE
    return { "jd_id": jd.id, "confidence": jd.confidence }


@router.get("/api/match/quick")
def quick_match(jd_id: str, user=Depends(current_user)):
    # 侧栏专用：把完整诊断裁成侧栏能放下的粒度
    profile = talent.get_profile(user.id)
    return diagnoser.quick(profile, graph.get_job(jd_id))

# crawler.py —— 后端批量通道，和插件完全独立
# 它不开扩展、不调扩展；扩展也不调它。两者只在 pipeline.ingest 汇合
from playwright.async_api import async_playwright
from apscheduler.schedulers.asyncio import AsyncIOScheduler

async def crawl_site(site, pages=5):
    async with async_playwright() as p:
        browser = await p.chromium.launch(headless=True)   # 服务器上无头跑
        page = await browser.new_page()
        for n in range(pages):
            await page.goto(f"{site.base_url}?page={n}")
            await page.wait_for_selector(site.list_selector)   # 等列表渲染完再取
            links = await page.eval_on_selector_all(
                site.item_selector, "els => els.map(e => e.href)"
            )
            for url in links:
                jd = await fetch_detail(page, url, site)        # 进详情页再抓一次
                pipeline.ingest(jd, source=f"crawler:{site.name}")   # ← 与插件同一个入口
            await asyncio.sleep(site.delay)                     # 限速：不触发频控、不绕验证码
        await browser.close()

# 定时：每天 02:00 回扫一轮存量岗位，产出写进同一张图谱
scheduler = AsyncIOScheduler()
scheduler.add_job(crawl_site, "cron", hour=2, args=[liepin])