agentleFS
eklavyagoyaleklavyagoyal/wingman/skills/job-search/SKILL.md

job-search

Skill0 starsChanged 2 months agoMITView on GitHub

Search German job boards for roles matching your CV and preferences

32 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use application-tracker, job-application and 2 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • xiding-codeSkill

    application-tracker

    Automatically log and track all job applications submitted. Use this skill after every application submission to record the details in a tracking spreadsheet and save relevant documents. Also trigger when the user asks "application status", "what have I applied to", "tracking", or wants to review their application history. Maintains a single source of truth for all job search activity.

    Also in 4 projects that use this

  • pepebot-spaceSkill

    job-application

    Write tailored cover letters and job applications using your CV and preferred style

    Also in 3 projects that use this

  • claude-office-skillsSkill

    company-research

    Conduct comprehensive company research and due diligence. Analyze business model, competitive landscape, management, and market position.

    Also in 3 projects that use this

  • claude-office-skillsSkill

    resume-tailor

    >

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 28 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read resume/*, preferences.md, profile.md, job-history.md (to skip duplicates), tracker.md, and contacts.csv if present.

    line 19
  2. Distinctive:

    Search terms from the argument, else target roles from preferences.

    line 23
  3. Very specific:

    Search German job titles. The board indexes by official Berufsbezeichnung, so Softwareentwickler finds far more than Software Engineer, and …

    line 45
  4. Distinctive:

    If the selector misses, screenshot to understand the layout, then write a site-specific selector.

    line 65
  5. Distinctive:

    Do not run the full A–H evaluation on every hit — that is expensive and most listings do not deserve it.

    line 73
  6. Distinctive:

    Never show the user an aggregator URL: If a direct link cannot be resolved, give the company and title so they can find it, and say the …

    line 100
  7. Distinctive:

    Do not tailor CVs, write letters, or fill forms in this skill.

    line 132
  8. Very specific:

    Update preferences.md from any feedback: "no Zeitarbeit" → dealbreaker.

    line 136

“Common” is measured: kept by at least 6 in 10 of the 28 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Job Search

Priority hierarchy: shared/references/priority-hierarchy.md · Boards: shared/references/de-job-boards.md · Interaction: shared/references/interaction.md

Searches the German market — including the boards US-built tools miss entirely.

Step 0: Prerequisites

Resolve the data directory. Check per shared/references/prerequisites.md: CV and preferences required, profile recommended.

Step 1: Load context

Read resume/*, preferences.md, profile.md, job-history.md (to skip duplicates), tracker.md, and contacts.csv if present.

Record the board with every job you log, in job-history.md and in any tracker row you create. It is free at this moment and unrecoverable later, and it is what patterns uses to tell the candidate which boards actually produce replies rather than listings.

Search terms from the argument, else target roles from preferences. For German-language searching, search both languages — „Softwareentwickler" and "Software Engineer" surface different employers, and the German term reaches the Mittelstand.

Step 2: Pick boards

Choose per profile rather than scanning everything. Ask with AskUserQuestion on the first run, then remember the answer in preferences.md:

ProfileBoards
DefaultArbeitsagentur, StepStone, LinkedIn, Indeed.de
Tech / startup+ join.com, BerlinStartupJobs, LinkedIn weighted up
Mittelstand / regional+ Personio feeds, Jobware, regional Arbeitsagentur filters
Öffentlicher Dienst+ Interamt, Bund.de
Academia / research+ academics.de, jobvector
Graduate / Werkstudent+ Absolventa, Get in IT

Arbeitsagentur is always in the set. It has the largest volume in Germany and carries employers that appear nowhere else — and it has a real JSON API, so it needs no browser at all:

node "${CLAUDE_PLUGIN_ROOT}/tools/arbeitsagentur.mjs" "Softwareentwickler" Karlsruhe --umkreis 30
node "${CLAUDE_PLUGIN_ROOT}/tools/arbeitsagentur.mjs" "Pflegefachkraft" München --json    # to post-process

Search German job titles. The board indexes by official Berufsbezeichnung, so Softwareentwickler finds far more than Software Engineer, and Pflegefachkraft more than Nurse. Run both if the role has a common English title in Germany.

It excludes Zeitarbeit and Personalvermittlung at source and tells you how many it hid. On a sample Karlsruhe software search that was 27 of 84 — a third of the visible market is intermediaries, so never present the filtered count as the whole market, and never present an agency listing as a direct role. --include-agencies shows them, labelled.

Where the API states a Gehaltsspanne, use it. German ads rarely name a salary, so a real range from the BA beats any researched estimate, and it belongs in Block E of the evaluation.

Step 3: Search

Browser tools per shared/references/web-extraction.md.

Never call get_page_text on a results page — it returns the whole document and can blow out the context window unrecoverably. Extract only listing rows with javascript_tool:

Array.from(document.querySelectorAll('[class*="job"], [class*="listing"], [class*="card"], article, [role="listitem"]'))
  .slice(0, 50)
  .map(el => el.innerText.trim())
  .filter(t => t.length > 20 && t.length < 500)
  .join('\n---\n')

If the selector misses, screenshot to understand the layout, then write a site-specific selector. Prefer structured feeds where they exist (Personio XML, the Arbeitsagentur Jobsuche API).

Cookie banners: decline non-essential. If a banner blocks reading entirely, say so — do not click through consent silently.

Do not retry a failing board more than once. Note it, move on, and tell the user which boards were skipped and why. A silently partial search is worse than a short one.

Step 4: Triage

Do not run the full A–H evaluation on every hit — that is expensive and most listings do not deserve it. Triage cheaply first:

  1. Drop anything already in job-history.md
  2. Label Zeitarbeit and Personalvermittlung per de-job-boards.md. Never present a leased-labour role as a direct one.
  3. Screen against dealbreakers and the salary floor from preferences.md
  4. Flag the language requirement of each ad, and whether it clears the candidate's level
  5. Flag work-authorization blockers per work-authorization.md — remembering that a permit holder is not blocked by a "no sponsorship" line

Rank what survives. Then run the full /wingman:evaluate only on the top handful, or on whatever the user picks.

Step 5: Log everything

Append every job seen to DATA_DIR/job-history.md — including rejects, with the reason, so the same listing is not re-surfaced weekly:

## [DATE] — Search: "[terms]" — Boards: [list]

| Titel | Firma | Ort | Gehalt | Sprache | Typ | Vorauswahl | Notiz |
|-------|-------|-----|--------|---------|-----|------------|-------|

Typ is Direkt / Zeitarbeit / Vermittlung. Sprache is the ad's language.

Step 6: Resolve employer URLs

For each surviving job, click through to the actual employer posting and capture that URL. Save the posting to DATA_DIR/jobs/[company-slug]-[date]/posting.md with the URL at the top.

Never show the user an aggregator URL. If a direct link cannot be resolved, give the company and title so they can find it, and say the link resolution failed.

Step 7: Present

Compact, and always with the German context the user needs to judge:

## [DATE] — [n] new matches from [m] boards

### 1. [Titel (m/w/d)] — [Firma]
- **Sprache**: German ad, C1 required → your level: B2 ⚠️
- **Vergütung**: 70.000–80.000 EUR (+13. Monatsgehalt) · 30 Tage Urlaub
- **Typ**: Direkt · unbefristet · Mittelstand, Stuttgart · Hybrid 2 Tage
- **kununu**: 3.8 — reviews mention long hours
- **Arbeitserlaubnis**: no issue (Blaue Karte held)
- **Netzwerk**: you know [Name] ([Position]) there
- **Warum**: [one line]
- **Anzeige**: [employer URL]

Omit any line with no data. Never pad.

Then note which boards were searched, which failed, and how many were filtered with the top reasons — so a thin result set is explainable rather than mysterious.

Step 8: Next steps

/wingman:evaluate <url>     full A–H report + 1–5 score
/wingman:tailor-cv <url>    Lebenslauf or CV
/wingman:apply <url>        evaluate, write, and fill the form

Do not tailor CVs, write letters, or fill forms in this skill. Those are separate skills with their own gates.

Step 9: Learn

Update preferences.md from any feedback: "no Zeitarbeit" → dealbreaker; "nothing below 70k" → floor; "prefer Berlin" → location weight.


Response Format

  1. Matches — ranked, with language, comp, type, kununu, authorization, network
  2. Search Coverage — boards searched, boards failed, counts filtered and why
  3. Next Steps

Permissions

{ "permissions": { "allow": [
  "Read(~/.wingman/**)", "Write(~/.wingman/**)", "Edit(~/.wingman/**)",
  "mcp__claude-in-chrome__*"
] } }

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.