nowcoder-public-intel
Collect, normalize, and synthesize public Nowcoder / 牛客 search, discuss, interview-experience, job, and offer-signal posts for career, recruiting, salary, JD, and interview research. Use when Codex ne
SKILL.mdskills/nowcoder-public-intel/SKILL.md
GitHubNowcoder Public Intel
Use this skill for low-volume public 牛客 / Nowcoder intelligence. Treat outputs as community signals, not authoritative facts.
Safety Defaults
- Use only public pages or unauthenticated public endpoints.
- Do not request, read, store, or paste cookies, tokens, headers, local browser profiles, usernames, contact data, private screenshots, or private comments.
- Do not bypass login, CAPTCHA, paywalls, anti-bot controls, hidden-content permissions, or rate limits.
- Keep results small. Prefer
--max-pages 1or2; avoid broad historical crawling. - Save only normalized metadata, links, short snippets, and your own synthesis.
- Mark offer/salary claims as low confidence unless cross-validated by official JD, multiple samples, or another channel.
Quick Commands
From a repo that has the skill installed globally:
$skill = "C:\Users\Administrator\.codex\skills\nowcoder-public-intel"
python "$skill\scripts\nowcoder_public_intel.py" schema
Search public Nowcoder:
$skill = "C:\Users\Administrator\.codex\skills\nowcoder-public-intel"
python "$skill\scripts\nowcoder_public_intel.py" search `
--query "腾讯 大模型 博士 offer 深圳" `
--tag 面经 `
--order create `
--max-pages 1 `
--max-results 10 `
--output nowcoder-results.jsonl
Batch search:
$skill = "C:\Users\Administrator\.codex\skills\nowcoder-public-intel"
python "$skill\scripts\nowcoder_public_intel.py" batch-search `
--query "华为 大模型 博士 offer" `
--query "腾讯 青云 博士 offer" `
--query "大模型 算法 博士 深圳" `
--max-pages 1 `
--max-results 10 `
--output nowcoder-batch.jsonl
When network fetches fail or access is blocked, use planning mode:
python "$skill\scripts\nowcoder_public_intel.py" plan `
--query "字节 大模型 博士 offer" `
--tag 面经
Workflow
- Scope the question.
- Identify companies, cities, job families, degree level, and timeframe.
- Prefer concrete query bundles such as
腾讯 大模型 博士 offer 深圳,华为 OD 博士 offer,大模型 推理 infra 面经.
- Run a small public search.
- Use
searchfor one query orbatch-searchfor a bounded set. - Use tags only when helpful:
面经,求职进度,内推,公司评价. - Use
--order createfor recency-sensitive scans. - Use
--max-resultsto cap per-query output before writing project artifacts.
- Normalize evidence.
- Keep
url,title,snippet,created_at, counts, platform, query, tag, confidence, and matched signal keywords. - Do not copy full post bodies unless the user explicitly asks for a short, compliant excerpt from public content.
- Keep
- Synthesize conservatively.
- Separate JD/interview signals from salary/offer signals.
- Label rumors, single posts, and ambiguous degree/level data as low confidence.
- Cross-check with official career pages, company programs, or other independent samples before using in a career-value conclusion.
- Record source ledgers.
- For project work, save JSONL/CSV/Markdown in the relevant project folder.
- Include access date and uncertainty notes.
Script Behavior
The bundled script first tries Nowcoder's public search endpoint. If that fails, it emits structured fallback URLs for manual/browser or AnySearch follow-up rather than switching to private/authenticated routes.
Read references/source-routing.md when choosing between public endpoint, page search, RSSHub-style routes, AnySearch fallback, and manual review.
Read references/output-schema.md before integrating results into a career-value matrix or source ledger.
Non-Goals
- No 脉脉, Offershow, BOSS, Liepin, Xiaohongshu, or Zhihu crawling.
- No login-required comments, private feeds, user-profile harvesting, contact extraction, automated messaging, or resume/job-application automation.
- No claims that Nowcoder posts prove salary levels by themselves.