Fighter90/career-ops-ui v1.270.0

on GitHub

[1.270.0] — 2026-11-07

Added

• Six new scan sources (registry 135 = 130 EN + 5 RU · 130 adapters) — ALDI Nord (single-company provider, per-tenant detail pages, full stop-reason machinery), HiBob (tenant boards <tenant>.careers.hibob.com, one job-ad API request), HireHive (<tenant>.hirehive.com v2 jobs API; later-page failures keep partials per the house contract), Factorial (tenant careers sites on the factorial domain family, CSP-derived allowlist, cross-domain tenant matching), Polymer (careers-API walk; RAW-length stops + a covering-count guard so a demo tenant's lying total cannot end a real board), Remote (apply.remote.com's own ATS; closed/not-accepting rows drop while unseen statuses never do — and region codes are kept AND spelled out so either vocabulary matches a location filter).
• Scanner reads whole boards: Consider walks its full catalog through the board's own meta.sequence with per-page retry and a hard cap warning (the ~10-posting first page was the whole scan); a country-only location filter now accepts worldwide labels (worldwide, remote (worldwide)) without widening a city list; a portals.yml entry can raise its request timeout with timeout_ms; CAREER_OPS_HTTP_TIMEOUT_MS is read per request, so setting it in .env after boot works; scan concurrency is configurable (CAREER_OPS_SCAN_CONCURRENCY / CAREER_OPS_SCAN_HOST_CONCURRENCY, defaults unchanged 32/10).
• Telegram strict source grows real-world parsing — Habr Career vacancy links (only /vacancies/<id> is a vacancy; profiles, lists and searches refuse), four positional employer shapes (Employer ищет …, Title в Employer, second-line Employer is a …, employer after the salary) with refusal rules for role/sector/department/format/person/place captures, and agency-withholding phrasing («название компании-заказчика не раскрыто») drops the post.
• Job descriptions reach the content filter on six more boards — RemoteOK, Remotive, We Work Remotely, Working Nomads, Jobspresso (the full post, not the WordPress excerpt) and Arbeitnow now carry a description; NoDesk titles decode the twice-escaped ampersand; Jobicy decodes HTML entities before deciding a row is empty.

Fixed

• Avature accepts query-style /JobDetail?jobId= links and cards whose article class is not the leading one (Deloitte USI); Arbeitsagentur keeps reference numbers containing a slash working end to end (outgoing URL double-encodes, the liveness rung decodes once and compares the real refnr); Getro warns when getro_max_pages cuts a board short — and stays silent when the board genuinely ended.
• Contributors list — goldbillka joined ×19 READMEs, ta/vi gained the avatar cards, release badges actualized.

Notes

• The board-body-cap fallbacks the parent added have no web-ui counterpart: this scanner's transport has no body-size cap (reads are bounded by the request budget, never a size failure) — documented no-port.
• The verify-portals probe-timeout fix is a no-port: the health surface probes careers_url pages, not ATS board APIs.
• Telegram strict keeps its Unicode-boundary LOCATIONISH_RE — the fork divergence is intact.