[rahulrathore95]

Summary

Changes made while using Dossier for my own job search, across two commits. Happy to split these into separate PRs if that's easier to review.

Fix: JSearch's 10 s timeout was shorter than its real-world latency. A live request against OpenWeb Ninja's JSearch API took 10.3 s to answer, so every search lost the race against src/lib/job-search/http.ts's hard-coded 10 s limit and came back "JSearch isn't answering right now" even with a valid key. Raised to 25 s. (Deliberately kept as a plain constant, not read from the environment — an earlier version of this fix read it from env at module scope, which forced full environment validation on every module that imports job-search code, breaking ~15 unrelated test suites and any code path that registers job types without a job source configured. Caught and reverted during testing.)

Feature: blocked companies (Settings → Job search). A list of companies that Find jobs, the job agent and the Claude connector's search_jobs/save_job_scores/save_application/draft_application never show, score or draft an application for again. Matched against a posting's company, title and description, so a staffing listing that only names the employer in the body text is still caught. Filtered centrally in askSource() and the job agent's search step, with a defense-in-depth check in the connector tools for postings reached by posting_id/posting_url/paste_id instead of a search.

Feature: Firecrawl, a fourth job source. Runs two web searches per page — one biased toward LinkedIn, Indeed, Naukri, Glassdoor, Monster, Greenhouse, Lever, Ashby, Workday and SmartRecruiters, one with no site restriction at all for whatever that fixed list misses (a smaller board, a company's own careers page) — merged into one result, closer to what a person googling "data engineer jobs in Bengaluru" would see than Dossier maintaining a scraper per site. Discovery-only: posting details are fetched with the existing guarded fetchPosting(), exactly like Adzuna's adapter. Needs a Firecrawl key (DOSSIER_FIRECRAWL_KEY or Settings → Job search) and one migration for the new PostingSource enum value.

Feature: falls back across job sources. searchWithFallback() tries the source chosen in Settings first, then every other source with a key saved (each source keeps its own key independently of which is "chosen"), in case the first is out of quota, down, or doesn't cover the query's country. A source with no key configured is skipped silently rather than counted as a failed attempt, so with only one source set up — the common case — the error message is unchanged from before this existed. A kept search's "Show more" always stays pinned to the exact source it started with, since its pagination cursor only means something there. Shared by askSource (Find jobs, search_jobs) and the job agent, which additionally excludes the Bundesagentur für Arbeit from its own fallback chain (it never searches that source unattended, #223).

Feature: hyperlinks on the resume. Email, website and LinkedIn in the header were plain text; they're clickable links now (mailto:, tel:, https://), styled the same muted, underline-free way a publication's citation link already was. ResumeContent.contacts changed shape (string[] → {text, href}[]) for this; the cover letter's own sender block, which is free text the user can edit, is untouched.

Test plan

• yarn typecheck, yarn lint, yarn test (full suite) all pass — run repeatedly for stability across both commits, 1717 tests, 0 failures
• The new Prisma migration (20261028120000_firecrawl_source) applies cleanly against a fresh and an existing database
• A production image built from this Dockerfile starts, passes its health check, and keeps existing data and sessions intact across the swap — verified against both a fresh database and a real one already in use
• Verified in a real browser against a live instance: chose Firecrawl as the source, saved and tested a key, added and removed blocked companies, confirmed persistence across a reload
• Resume snapshots regenerated and reviewed by hand (only the new <a> wrapping changed, nothing else moved)
• New unit/integration tests: the blocklist matcher, the Firecrawl adapter (including its two-search merge), the fallback chain (single-source behavior unchanged, falls back on quota, combined message when everything fails, "Show more" stays pinned), and the linked contact fields — plus updates to every existing test the new fields/sources touch

🤖 Generated with Claude Code

[AienTech]

Hi, thanks for the contribution. I'll take a look at it and will let you know. so far, having separate PRs for this is better. since JSearch's timeout is already being released today.

new comment added

[AienTech]

Hi @rahulrathore95, thank you for your contributions. I know Dossier is still missing features and has bugs, and I appreciate every contribution, including ones made with AI. But I can't accept a single PR this size that I can't review properly or test with confidence.

Please split it into smaller PRs, one change each, small enough for a person to read and test:

• blocked companies
• Firecrawl as a job source
• falling back across job sources
• hyperlinks on the resume

The JSearch timeout fix is covered by today's release, so you can leave that one out. For bigger features like Firecrawl, please open an issue first so we can agree on the approach before you write the code.

I'm closing this PR for now. Thanks again for taking the time.