Company skip list and profile deal-breakers now ignore hyphens when matching,
apply skip keywords to job titles as well as employers, and short-circuit
scoring when a deal-breaker hits so co-op posts cannot score 80 via the LLM.
Co-authored-by: Cursor <cursoragent@cursor.com>
The jobs.json feed often labels roles Remote/Worldwide while the public
job page JSON-LD and description include the real city (e.g. Mumbai/Nagpur).
Enrich vague rows by reading each QAJobsBoard detail URL before import.
Co-authored-by: Cursor <cursoragent@cursor.com>
QAJobsBoard and similar feeds often store Worldwide/Remote while the real
country is only in the description. Scan title and description when location
is vague, and prefer concrete locations from QAJobsBoard postings.
Co-authored-by: Cursor <cursoragent@cursor.com>
Reject hybrid or partial-office postings at ingest so the Remote badge and
filters match fully remote roles. Cron can PATCH search geography, remote-only
workplace types, and QA search terms before each scheduled pipeline run.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add blockedCountries in Settings so pipeline discovery drops jobs whose
location mentions listed countries (existing discovered rows are kept).
Document the feature, fix smoke tsconfig inheritance for nested extractors,
and run smoke via an absolute-tsconfig wrapper.
Co-authored-by: Cursor <cursoragent@cursor.com>
Adds Arc.dev, BC T-Net, Eluta, iCIMS tenants, QAJobsBoard, and SmartRecruiters
manifests with registry/settings/UI wiring; registers full extractor list in
smoke-extractors and documents supplementary board access paths. Aligns Careerjet
v4 with the url query parameter and fixes strict typing in QAJobsBoard.
Co-authored-by: Cursor <cursoragent@cursor.com>
The Dockerfile only copied a fixed list of extractor dirs; new sources
were listed in shared but their manifest.ts files were absent from the
container, so discovery logged missingManifest in production.
Copy extractors/ once before npm install in builder and production, and
skip redundant per-extractor COPY lines. Add extractors/*/storage/ to
.dockerignore to avoid baking local cache into the build context.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add shared helper to parse blocked company keywords from JSON or legacy text
- Wire discover-jobs to resolved keywords for employer blocking; extend filters UI
- Include vite/client in TS types for import.meta.env; fix JobListItem test fixture
Made-with: Cursor
- Add per-listing notes (JobNotes component, notes column, auto-save)
- Enforce hard score cap (≤15) when deal-breaker hits are present;
add clearance/citizenship deal-breaker rules to scoring prompt
- Cover letter prompt now uses full search profile (experience level,
skills, work arrangement, location, salary, industries, deal-breakers)
and produces longer, name-signed output
- Include candidate name + location in sanitized scorer profile
- Raise MAX_JOB_ACTION_BATCH_SIZE from 100 → 2500 (shared constant)
- Update README with new features
Made-with: Cursor
Use JOBOPS_LOCAL_RESUME_PATH or Settings local path as the tailored PDF
source when Reactive Resume template is not selected. Align GET
/api/profile/status with local resume configuration. Update route test
mocks for resolveLocalResumeFilePath.
Includes discovered-panel UI tweaks in DecideMode and DiscoveredPanel.
Made-with: Cursor
- Put match badges in their own flex-wrap row under the title (no run-on text)
- Toolbar: flex-wrap, min-w-0 label, shrink-0 count, middot separator on narrow layouts
Made-with: Cursor
- Always show Cover letter card with Generate when empty; Regenerate + Copy when present
- Shared useCoverLetterGeneration hook calling runJobAction(generate_cover_letter)
- Ready tab previously had no cover letter UI; add same block under Fit Assessment
- Floating bar Cover letter button tooltip points to Overview/Ready
- Job detail ⋯ menu: Generate / Regenerate cover letter
Made-with: Cursor
- Require workArrangementMatch in LLM schema; show badge in Fit Assessment
- Prompt: remote-only vs hybrid/N days in office is a hard cap (~35 score)
- Deterministic cap from JD keywords + isRemote/workFromHomeType when only Remote is selected
- hasNonEmptyProfile includes preferredWorkArrangement and preferredLocations so prefs reach the scorer
Made-with: Cursor
- Cap suitability score when roleTypeMatch is low and target roles are set
- Harden scoring prompt (role-type gating, examples, headline in resume payload)
- QA-oriented mock scoring when LLM is unavailable
- .env.example: optional Ollama MODEL/LLM_BASE_URL block for self-hosted LLM
Made-with: Cursor
- Add glassdoor to DEFAULT_CONFIG so POST /api/pipeline/run with {} runs JobSpy trio.
- jobber-pipeline-telegram.sh: optional JOBBER_PIPELINE_SOURCES for explicit source list.
- Deploy doc: WAL checkpoint + rsync jobs.db to VM for profiles/settings/jobs.
Made-with: Cursor
- Add search profiles (DB, API, settings UI) and wire into scorer/pipeline search terms.
- Add cover letter generation (service, job action, JobDetail UI).
- Align JobSpy Indeed country with country-level search geography when settings conflict; warn in logs.
- Infer country from search cities via inferCountryKeyFromSearchGeography (shared).
- Ignore extractor venv/storage and local data in Biome; ignore orchestrator/storage and JobSpy .venv in git.
- Vite: do not watch orchestrator/storage (prevents reloads during startup.jobs pipeline).
- JobSpy: document Python 3.10+ and venv setup in README/requirements.
- Onboarding and settings: local resume path handling, orchestrator .env.example for Vite.
Made-with: Cursor
Production images previously omitted the flag, so the full RxResume wizard
always appeared. Default skip matches self-hosted LLM-in-.env workflows;
override via docker-compose build args if needed.
Made-with: Cursor
- Normalize SQLite vs ISO discoveredAt for run-window filtering
- Retry jobs list after run; compressed JSON requests
- Cross-check GET /api/jobs/revision when list is empty
- Docs: cron at 9/13/18, pull + redeploy steps for VM
Made-with: Cursor
- Add remote/orchestrator filter by workplace (remote, not remote, unknown) with URL param
- Expose isRemote on job list API; canonicalize URLs and source_job_id dedup on import
- Onboarding: optional VITE_SKIP_RXRESUME_ONBOARDING for RxResume-free onboarding
- Scoring UI + docs for company skip list; pipeline-run dedup note
- Vitest: TZ=UTC for stable time-based tests
- DEPLOY_GITEA_VM_CRON_TELEGRAM.md for VM/cron/Telegram ops
Made-with: Cursor
Turn Ghostwriter's flat message list into a tree structure, enabling
Claude.ai/ChatGPT-style branching conversations.
**Data model**: Add `parentMessageId` and `activeChildId` to messages,
`activeRootMessageId` to threads. Migration backfills existing messages
into a linear chain and links regenerated messages as siblings.
**Backend**: Tree-walking queries (getActivePathFromRoot, getAncestorPath,
getSiblingsOf), rewritten history builder that follows the ancestor path,
new editMessage and switchBranch services, regenerate now works on any
assistant message (not just the latest).
**Frontend**: BranchNavigator component (← 2/3 → arrows), inline edit on
user messages, per-message regenerate on assistant messages, regenerate
button removed from composer (now per-message).
**Infra**: Pin Node 22 via Volta to prevent ABI mismatches with
better-sqlite3 across environments.
When switching between job tabs (e.g. Discovered → Applied), the detail
panel would briefly show the previously selected job with the new tab's
action buttons — confusing and error-prone.
Three coordinated fixes:
- setActiveTab now checks if the selected job's status fits the target
tab, keeping it when valid (e.g. Discovered → All Jobs) and clearing
it otherwise.
- New visibleSelectedJob useMemo guard synchronously nulls out the
selected job when it doesn't belong to the active tab, eliminating
the one-frame flash caused by the data hook's useEffect lag.
- Auto-select effect now handles the case where a job passes the status
check but gets filtered out by source/sponsor/salary filters.
* Add startup.jobs extractor support
* Harden startup.jobs extractor inputs
* Wire startupjobs into Docker and CI
* Tighten startupjobs review follow-ups
* fix: publish ghcr during release workflow
* feat: add startupjobs max jobs configuration and update related tests
