santifer/career-ops

Time: ~2-3h · Difficulty: Intermediate · Area: scanner What Add a domain-level aggregator/repost detection list to the scanner, complementing the existing SimHash cross-listing warning (FAQ §5). Today the scanner catches agency reposts by content: a SimHash fingerprint of each job description flags near-identical duplicates. A domain list catches them by source, which is cheaper, works when the text has been rewritten, and needs no fingerprint at all. They're complements, not competitors. Seed (with credit) @gup-abhi enumerated ~20 aggregator boards that repost other people's listings in #1944 (Recruit.net, Zippia, Expertini, Jooble, Adzuna, Neuvoo, JobisJob, Jobrapido, …), including the hard part most lists get wrong: Indeed is carved out as a primary board, not a scraper, because companies post there directly. That list is the starting point, and landing it earns the credit it deserves: keep the carve-out reasoning in a short note per entry. Acceptance criteria A data file (e.

Time: ~20 min · Difficulty: Beginner Several user-facing scripts documented in docs/SCRIPTS.md still have no --help, so running one with a wrong flag gives you no way to discover the right one. We already fixed this for stats.mjs, followup-cadence.mjs, detect-reposts.mjs and check-liveness.mjs; these four are the ones left: build-dashboard.mjs cv-sync-check.mjs normalize-statuses.mjs plugin-install.mjs What to do Pick ONE script from the list above. Add a USAGE string and a KNOWN_FLAGS array, and print USAGE and exit 0 when --help or -h is passed. Unknown flags should exit non-zero with the usage block, rather than being silently ignored. Document the real flags the script already accepts: read the argument handling first, do not invent options. Acceptance node <script>.mjs --help prints the usage block and exits 0. node <script>.mjs --nonsense prints the usage block and exits non-zero. No behaviour change when the script runs normally. One script per PR: that keeps review ea

Time: ~25 min · Difficulty: Beginner · No-code · Native speaker of the target language needed The interview skills live in modes/interview/ (English) and are translated per language under modes/{lang}/interview/. Five languages already have plan.md, practice.md and debrief.md translated, but are still missing the folder's README.md: the index page a reader lands on first, so right now those readers hit an English page in an otherwise translated folder. Missing in: de, id, ru, ua, zh. What to do Pick ONE language you speak natively: de, id, ru, ua or zh. Translate modes/interview/README.md into that language and save it as modes/{lang}/interview/README.md. Mirror the structure exactly: same headings, same table columns, same order. Only the prose changes. Leave file names, paths and links untouched (plan.md, ../interview-prep.md, config/profile.yml and the rest stay as they are). Acceptance Natural prose in the target language, not machine-literal. Structure matches the English or

Time: ~60-90 min · Difficulty: Beginner · Docs only, no code docs/RUNNING_ON_A_BUDGET.md section 5 ("Local LLM Tradeoffs") tells you what not to do: don't use 8B models, here's the VRAM you need, and a tip that a cheap hosted API is usually better value. All of that is true and useful. What it never gives you is the actual steps. Section 3 does exactly that for hosted setups, in a format that works: the "Kimi K2.5 via OpenCode (Verified)" recipe (line ~127) is numbered, copy-pasteable, names the config file, and is honest about what was and wasn't tested. Section 5 has no equivalent, so a reader who has the hardware and wants to go local has to figure it out alone. What to add One verified quick-start recipe for running a local model end to end, as a new subsection of section 5, mirroring the Kimi recipe's shape. Ollama through OpenCode is the natural first one, since section 3 already says OpenCode routes to local endpoints but never shows how. LM Studio would be just as welcome and

Time: ~2-4h · Difficulty: Beginner · Requires: native/fluent Portuguese (Brazil) — reserved for first-time contributors The interview modes (modes/interview/plan.md, practice.md, debrief.md) exist in several locales but not yet in Portuguese (Brazil) (modes/pt/ has the evaluation modes, not interview/). The exact pattern to mirror: PR #3212 (Ukrainian). Three things it got right that made it merge without a single change requested: Keep [inferred] and [inferred from JD] as English machine-readable tokens, with a Portuguese (Brazil) gloss alongside if helpful Register modes/pt/interview/ in the SYSTEM_PATHS list of update-system.mjs (one line, mirror the modes/ru/interview/ entry) Translate the anti-fabrication rules faithfully: the prohibition is absolute and admits no qualifier Acceptance: the three files under modes/pt/interview/, the SYSTEM_PATHS line, and node test-all.mjs green. Comment /assign below to claim this issue.