job.pro

Query Chinese big-tech campus + social recruiting from your terminal — job.ha7ch.com

npx @ha7ch/job-pro@latest tencent search "后台开发"
npx @ha7ch/job-pro@latest bytedance search "后台开发" --scope social

No signup, no token, no proxy server. 50 companies, all live. The CLI talks straight to each company's public API (e.g. join.qq.com for Tencent) and prints JSON. Pipe it into jq, Claude Code, anything.

Campus + social (1.1.0+)

Every adapter accepts a unified --scope flag:

job-pro bytedance search "后台开发" --scope social --page-size 5
job-pro xiaohongshu all                --scope campus
job-pro find "AI infra"                --scope social --text

--scope works on search, all, match, and the cross-company find verb. Default (flag omitted) preserves each adapter's 1.0.93 behaviour bit-for-bit — usually campus or a mixed feed. Adapters that can't query the requested channel (Tencent / JD / Cainiao / WeBank / Hikvision / CICC / Unitree have no public social-hire API) refuse fast with a useful message instead of silently returning empty. On find, those companies are silently skipped and reported under companies_skipped_by_scope.

Run job-pro help for the full company list, or see the roadmap matrix at job.ha7ch.com. Per-release notes: CHANGELOG.md.

Demo: hand it to Claude Code

Drop the prompt from job.ha7ch.com into Claude Code, attach your resume, and let the agent drive the CLI end-to-end.

1. It pulls the city's intern list and shortlists roles against your resume.

2. It pulls multiple JDs in parallel and grades each one line-by-line.

3. It hands you a final verdict — apply, fall back, or skip.

Install

npm i -g @ha7ch/job-pro
job-pro --version
job-pro selftest          # 3-stage end-to-end check; ~3s

Or one-shot via npx:

npx @ha7ch/job-pro@latest help

What you can do today

# search & inspect jobs
job-pro tencent search "数据科学" --page-size 10
job-pro tencent detail 1200791473415778304
job-pro tencent all --page-size 100             # drain every open post

# announcements
job-pro tencent notices
job-pro tencent notice 284
job-pro tencent flow "腾讯2026实习什么时候开始" --question-time 2026-05-13

# resume tooling (all offline)
echo "..." | job-pro tencent match -
job-pro tencent resume-check resume.md

# local memory for tracking your hunt
job-pro tencent memory set "stack=Go,Python" "target_city=深圳"
job-pro tencent memory event applied "腾讯后台 1200791473415778304"
job-pro tencent memory list

# list / browse adapters
job-pro list                 # human-readable, grouped by ATS family
job-pro list --compact       # JSON for piping

# cross-company parallel search
job-pro find "intern" --text                       # scan all 50 in parallel
job-pro find "AI" --apply-ready --text             # only show buckets you can fire today
job-pro find "前端" --companies tencent,bytedance,alibaba --limit 5

Add --compact to any command for a single-line JSON output (pipe-friendly).

Phase 2 — auto-apply (1.0+)

Submit applications from the CLI. 50 / 50 adapters expose an application schema; 45 / 50 have an executor wired (3 anon Greenhouse/Lever

• 42 session-required); 45 / 50 have endpoint-verified status. The 5 external adapters (Liepin × 4 + Unitree WeChat) are structurally IM-mediated and intentionally stay browser / chat handoff only.

# 1. Set up your profile (one-time)
job-pro profile init --interactive   # prompts for the 5 essential fields (validates each)
job-pro profile lint                 # check format of every field (exit 1 on fail)
# or: job-pro profile init && $EDITOR ~/.jobpro/profile.json

# 2. Inspect a job's schema (no profile required)
job-pro xpeng apply 8548990002 --schema

# 3. Dry-run an application (no network)
job-pro xpeng apply 8548990002       # stage + preview the POST
job-pro xpeng apply 8548990002 --interactive --remember  # fills answers, persists to profile

# 4. Verify the wire format against an echo server (no upstream impact)
job-pro xpeng apply 8548990002 --debug-submit-to https://httpbin.org/post

# 5. Actually submit
job-pro xpeng apply 8548990002 --confirm-submit

# Script mode still exists when you intentionally do not want a prompt:
JOB_PRO_I_UNDERSTAND_REAL_SUBMIT=yes \
  job-pro xpeng apply 8548990002 --really-submit

Greenhouse / Lever boards (xpeng / hoyoverse / weride) submit anonymously. Every other family needs a captured session — run job-pro extension for the path + install walkthrough, log into the careers site once, click Export, then drop ~/Downloads/jobpro/<adapter>.session.json under ~/.jobpro/.

# After capturing nio's session via the extension:
job-pro nio apply 7639693860494543167 --confirm-submit

The CLI gates real submission behind four layers (refuses with a clear JSON mode at the first that fails):

1. User consent: either --confirm-submit interactive confirmation, or JOB_PRO_I_UNDERSTAND_REAL_SUBMIT=yes with --really-submit for scripts.
2. staged.ready — every required field is filled.
3. endpoint_verified === true (URL probe-confirmed or end-to-end smoked) — bypass with JOB_PRO_ALLOW_SPECULATIVE_ENDPOINT=yes.
4. For non-anon families: captured ~/.jobpro/<co>.session.json AND <30 days old — bypass staleness with --allow-stale-session.

Bulk-stage with apply --batch <file|- (newline-separated post_ids; - reads stdin); pairs with --form-file for uniform answers. Batch intentionally refuses real submission — verify wire format with --debug-submit-to, then submit jobs individually with --confirm-submit.

Five adapters (hikvision / cicc / cainiao / webank / unitree) are intentionally external — recruiting is mediated through Liepin recruiter chat or WeChat mini-programs; no API submission exists. For those, apply surfaces the apply_url to open in browser.

See docs/auto-apply.md for the full per-adapter submission flow matrix.

Roadmap

Phase 1 — Read jobs: 50 / 50 companies, all live. See the full live matrix with per-company status icons at job.ha7ch.com, or run job-pro help for the canonical list.

Coverage by source family:

Source family	Companies	Notes
Bespoke per-company API	23	Tencent, ByteDance, Alibaba, Meituan, Xiaohongshu, JD, …
Feishu Recruiting (ATSX)	9	Xiaomi, NIO, MiniMax, Zhipu, iQIYI, Agibot, Lilith via CDP, 01.AI, Baichuan
Moka (app.mokahr.com)	7	Moonshot, Megvii, DeepSeek, GalaxyUniversal, StepFun, Cambricon, Geely
Beisen Wecruit	2	SenseTime, Horizon Robotics
Beisen iTalent (zhiye)	2	vivo, iFlytek
Greenhouse / Lever	3	XPeng, WeRide, HoYoverse — international/US arms
Liepin third-party feed	4	Hikvision, CICC, Cainiao, WeBank (no canonical public feed exists)

23 + 9 + 7 + 2 + 2 + 3 + 4 = 50.

Phase 2 — Auto-apply is live: 45 / 50 adapters have a submitter wired (3 anon Greenhouse/Lever, 42 via captured browser session); the remaining 5 are intentionally external (Liepin recruiter chat × 4 + Unitree WeChat QR × 1 — IM-mediated, no API submission exists).

Notes on coverage edge cases

• Greenhouse / Lever boards (XPeng / WeRide / HoYoverse) only carry the international arm's postings (US AI center, Singapore game-dev, etc.). The China-side campus boards for these companies aren't publicly reachable from outside their networks; when they become accessible a sibling adapter will land.
• Lilith uses a Feishu tenant gated by a ByteDance Tengine _signature anti-bot token. The CLI cracks it via puppeteer-core driving the user's local Chrome. If Chrome isn't installed, this one adapter returns a helpful ok:false with the install hint — the other 49 are unaffected.
• Hikvision / CICC / Cainiao / WeBank have no canonical anonymous public feed (the first three are geo-fenced or DNS-internal; WeBank is WeChat- mini-program-only). For these four the CLI surfaces real currently-open positions through Liepin and clearly labels the result with source: "api-c.liepin.com" and attribution: "via Liepin (third-party aggregator) — official portal not publicly accessible". See docs/stub-unblock.md for the reasoning.

How it's built

• cli/ — the npm package (TypeScript, Node 18+). Single runtime dep: puppeteer-core (used only by the lilith adapter, see above).
• cli/src/<company>.ts — one thin adapter per company.
• cli/src/{feishu,greenhouse,lever,moka,wecruit,liepin}.ts — generic SaaS-ATS factories. Adding a new tenant on an existing ATS is a ~30-line wrapper.
• cli/src/cdp.ts — singleton headless-Chrome helper for anti-bot upstreams. Reads $JOB_PRO_HTTPS_PROXY for a CN-egress proxy when needed.
• cli/src/adapter.ts — the explicit CompanyAdapter contract every adapter must satisfy.
• cli/test/smoke.ts — strict gate: any live adapter regressing to ok:false FAILs the suite. KNOWN_LIMITED is currently the empty set.
• src/ — the job.ha7ch.com landing page (Next.js).
• python-reference/ — the original Python port for join.qq.com.
• docs/ — endpoint inventories per company, plus stub-unblock.md with the full recon history.
• extension/ — manifest v3 Chrome extension that captures careers-site session cookies + CSRF headers for Phase 2 auto-apply. Load it via chrome://extensions/ → Developer mode → Load unpacked. See extension/README.md.

Why "local-direct" instead of a hosted backend

The data is public. We don't store anything on a server, don't see your queries, don't rate-limit you, can't go down. The flipside: you get the upstream's quirks (typos in field names, etc.) — we paper over them in the client, but if the upstream changes, the CLI may need a release.

Credit

The endpoint inventory for join.qq.com was recovered by inspecting the official Tencent WorkBuddy skill bundle. We re-implemented the client in both Python and TypeScript with our own structure, naming, and matching heuristics. No prompt copy, no documentation copy, no skill body is reused.

Contributing

Adding a new company is mechanical:

1. Find its public listing/detail API. DevTools → Network on the careers site is the fast path, but for SPAs with anti-bot challenges (Tengine _signature, EdgeOne JS cookies, etc.) you may need cli/probe/<company>-network.ts running puppeteer-core to intercept the real XHR — see existing probes for templates.
2. Identify the SaaS ATS family. If it's already supported (feishu / greenhouse / lever / moka / wecruit / liepin), add a ~30-line wrapper that calls createAdapter({ … }). Otherwise write a bespoke adapter mirroring tencent.ts.
3. Wire it into cli/src/index.ts ADAPTERS and cli/test/smoke.ts. The satisfies CompanyAdapter clause will refuse to compile if any of the 9 required verbs is missing.
4. Add an entry to src/app/page.tsx's COMPANIES array.
5. Run pnpm test. The smoke gate runs every adapter in parallel and FAILs on ok:false for non-KNOWN_LIMITED adapters.
6. Open a PR.

The auto-apply phase needs more thought — see the roadmap doc.

License

MIT — see LICENSE.