scrape

Run the scrape workflow, which discovers every installed portal-search CLI and queries them in parallel using your configured search terms. Results are deduplicated against previous runs and presented as a fit-sorted table of new postings.

Each posting gets a canonical key generated by the job_key.py helper from company, title, and URL, stored in seen_jobs.json. Later runs skip any posting whose URL or company-plus-title combination already exists in that file or in the application tracker.

A built-in health check inspects each portal's output for null fields or zero yields, then runs one bounded sentinel probe using the portal's documented test query. Suspect portals are reported as degraded, broken, or inconclusive, and can be disabled with user confirmation.

Yes. If bun is unavailable, the workflow falls back to WebSearch using site-specific query strings for each portal. Fallback results are tagged as websearch-sourced so you know which rows may come from a stale search index.

A posting requiring a language not declared in your profile is marked low match regardless of skill fit. A declared language required at a higher level than yours is flagged as a red flag in the results without being auto-excluded.

No. For high and medium fit jobs it only generates LinkedIn people-search URLs for recruiters and team peers that you open yourself. It never scrapes LinkedIn search pages or fabricates contact names.