job-scraper

name	job-scraper
description	Searches the Australian job market (SEEK via the seek-search CLI, plus startup boards via web search) for new positions matching your profile. Deduplicates across runs. Triggers on: job scrape, find jobs, search jobs, new jobs, job search, scrape jobs, /scrape
allowed-tools	Read, Write, Edit, Glob, Grep, Bash, WebFetch, WebSearch, Agent, AskUserQuestion

Job Scraper (Australian market)

---

How It Works

This skill discovers Australian job postings using two channels:

1. SEEK — primary. The seek-search CLI (tools/seek-search/seek_search.py) queries SEEK's JSON API directly and returns structured jobs (title, company, location, salary, remote/hybrid, teaser, URL). SEEK is the largest Australian board and its HTML pages are Cloudflare-blocked, so the CLI is the only reliable way in.
2. Startup / ATS boards — secondary. WebSearch over Wellfound, Y Combinator, and Greenhouse/Lever/Ashby for founding-engineer and AI-startup roles SEEK under-indexes. (WebSearch is US-region, so treat its AU results as supplementary and verify AU-eligibility before presenting.)

It then deduplicates against previously seen jobs and the application tracker, and presents new matches with a quick fit assessment.

Invocation

Triggered by: "Find new jobs", "Scrape for jobs", "Any new positions?", "/scrape".

Optional arguments:

• A focus area, e.g. "/scrape ai" or "/scrape full stack"
• "broad" to run all priority categories
• "remote" to hard-filter to remote/hybrid roles
• "linkedin" to ALSO query LinkedIn (opt-in only — see Step 2b; off by default for ToS reasons)

---

Execution Steps

Step 0: Load State

1. Read job_scraper/seen_jobs.json (create if missing — start with {"seen": {}})
2. Read job_search_tracker.csv to extract already-applied companies+roles
3. Read search-queries.md (this directory) for the priority categories, role keywords, and --where locations
4. Read the candidate profile (.claude/skills/job-application-assistant/01-candidate-profile.md and 04-job-evaluation.md) to ground the fit assessment in Step 4

Step 1: Search SEEK (primary)

For each priority category in search-queries.md, run the CLI once per role-title keyword. By default run the top 3 priority categories; if the user said "broad", run all. If the user gave a focus area, prioritise that category's keywords.

cd tools/seek-search
python3 seek_search.py --keywords "<role title>" --where "<location>" --pages 2
# add --remote when the user asked for remote, or for an Australia-wide remote sweep:
python3 seek_search.py --keywords "<role title>" --where "All Australia" --remote --pages 2
# add --days N to limit to recent postings (e.g. --days 14 on a repeat/weekly run):
python3 seek_search.py --keywords "<role title>" --where "<location>" --days 14 --pages 2

• Run the CLI via the Bash tool. It prints a JSON array — parse it directly.
• Use the candidate's configured location tiers from search-queries.md for --where.
• Batch the keyword calls; each is fast.

Step 2: Search startup boards (secondary, optional)

For founding-engineer / AI-startup coverage, run a few WebSearch queries over wellfound.com, workatastartup.com, and job-boards.greenhouse.io. Only fetch a posting with WebFetch if it looks like a strong match AND is not a SEEK URL (SEEK 403s). Verify Australia-eligibility before presenting — many board results are US-only.

Step 2b: LinkedIn (OPT-IN ONLY — never run by default)

Only run this if the user explicitly opted in (said "linkedin", "/scrape linkedin", or asked for LinkedIn). Do not query LinkedIn otherwise — automating it is against LinkedIn's ToS, so it is off by default. When opted in, briefly remind the user it's at-their-own-risk, then:

cd tools/linkedin-search
python3 linkedin_search.py --keywords "<role title>" --where "Australia" --days 30

Keep volume low (a few keyword calls). Merge LinkedIn results into the same dedup/ranking as SEEK. For a full description or /apply, use linkedin_search.py --detail <id|url>.

Step 3: Deduplicate

For every job from Steps 1–2:

• Skip if its SEEK id/URL or company+title key already exists in seen_jobs.json.
• Skip if the company+role already appears in job_search_tracker.csv.

Step 4: Quick Fit Assessment

For each NEW job, do a rapid fit check against the candidate profile (NOT the full 04-job-evaluation.md pass — just a signal). Use the teaser, title, salary, work_arrangement, and bullet_points the CLI returned:

• High — role directly hits the candidate's core skills AND location/remote works AND (if salary shown) it meets the candidate's salary expectation.
• Medium — adjacent role, or location/salary needs checking.
• Low — significant skill gap, or on-site outside the candidate's city with no remote.

Apply the location filter from search-queries.md.

Step 5: Store

Add ALL surfaced jobs (new and skipped) to seen_jobs.json:

{
  "seen": {
    "<seek_id_or_company_title_key>": {
      "title": "...", "company": "...", "location": "...", "url": "...",
      "salary": "...", "work_arrangement": "...",
      "first_seen": "YYYY-MM-DD", "fit": "high/medium/low", "status": "new/skipped/evaluated"
    }
  }
}

Step 6: Present Results

Present new jobs in a table sorted by fit (high first):

## New Job Matches — YYYY-MM-DD

Found X new positions (Y high, Z medium, W low match).

| # | Fit | Title | Company | Location | Arrangement | Salary | URL |
|---|-----|-------|---------|----------|-------------|--------|-----|
| 1 | High | ... | ... | ... | Remote/Hybrid | ... | [Link](...) |

### High-Match Highlights
For each high-match job, add 2-3 bullets: why it matches, key requirements to check, any red flags.

After presenting, ask:

"Want me to evaluate any of these in detail, or apply to one? Give me the number(s)."

To evaluate or apply to a SEEK result, fetch its full description first (the search step only returns a teaser):

cd tools/seek-search && python3 seek_search.py --detail "<seek url or id>"

Then run the job-application-assistant workflow (or /apply) on that full description.

Step 7: Update Tracker (Optional)

If the user decides to apply, add a row to job_search_tracker.csv.

---

Important Rules

1. Never fabricate job postings. Only present jobs returned by the CLI or actual web results.
2. Respect deduplication. Always check seen_jobs.json AND job_search_tracker.csv first.
3. Location filter. Honour the candidate's configured location tiers; skip on-site-only roles outside their city unless the user opts in.
4. Only open positions. The CLI returns live listings; still skip anything clearly stale.
5. Full SEEK descriptions ARE fetchable via seek_search.py --detail <id|url> (GraphQL). The search step returns only a teaser; pull the full body with --detail before a full fit eval or /apply. No manual paste needed for SEEK URLs.
6. Efficiency. Batch CLI keyword calls; don't WebFetch every result — pre-filter on title/teaser/salary first.