Kowork collect

KOWORK Collect

Collect recent public postings from KOWORK's English jobs-for-foreigners listing and return normalized JSON (docs/SCHEMA.md). Self-contained: this skill's collect.py carries its own fetch + parse and depends only on curl_cffi, beautifulsoup4, lxml.

Boundaries

• Read-only. GET of the public /en listing only. Never log in, save a job, message, or apply. Do not submit any form.
• No fabrication. Fields the feed does not expose are null; titles, companies, locations, and visa/type tags are copied verbatim.
• A collected posting reflects the public page at run time; it is not proof the role is still open.

Run

If collect.py fails with ModuleNotFoundError, install the deps yourself (pip install curl_cffi beautifulsoup4 lxml) and retry — do not ask the user to install anything.

pip install curl_cffi beautifulsoup4 lxml   # or: pip install -r requirements.txt

python3 skills/kowork-collect/collect.py --query "developer" --limit 20
python3 skills/kowork-collect/collect.py --limit 20      # whole first page

• --query, -q — filter the listing by title/company substring (client-side; see Notes). Omit to return the whole first page.
• --limit, -n — max postings (default 20).
• --pretty — indent the JSON.

Output: a JSON array on stdout; a one-line count on stderr. KOWORK exposes company (from the logo alt), location, employment type, and metadata pills (visa eligibility like E-7 Sponsors, job category) — the full pill set is kept in snippet, with a deadline D-<n> in deadline when shown.

Notes

• Source: SRC-JOB-KOWORK — https://kowork.kr/en (collectible_open, re-verified 2026-07-07). This is a board for foreigners in Korea, so postings carry visa tags (E-7, F-series, etc.).
• First page only, best-effort. KOWORK is a server-rendered Next.js app; the static HTML holds ~15 postings (the first page). More requires a real browser (infinite scroll), which is out of scope for a plain-fetch collector.
• Keyword filtering is client-side. The ?search= param is ignored server-side, so --query filters the fetched page by title/company substring.
• Each card is an a[href*='/en/post/<id>']; external_id is that numeric id. Canonical detail URL: /en/post/<id>. Company comes from the logo image alt (job-list-<company>-logo); the metadata pills are duplicated across desktop/mobile blocks and are deduped on text.
• If the Tailwind markup changes and 0 postings return, re-derive the card anchor, p.line-clamp-2 title, and pill selectors against a live fetch (do not hardcode a specific company).