jobpilot
Top-level overview of the JobPilot automated job-hunt pipeline. Scrapes ~10 job sources via Apify, scores postings against your resume, writes a CSV report, tailors resumes for top matches, and pushes a Telegram digest on a schedule.
Resources
15Install
npx skillscat add ashlesh-t/jobpilot Install via the SkillsCat registry.
JobPilot
JobPilot is a personal, automated job-hunting pipeline for Claude Code. It scrapes job
boards via Apify, scores each posting against your resume, generates a CSV report, tailors
resumes for the strongest matches, and pushes a Telegram notification on a scheduled cadence.
Two-layer architecture (token strategy)
JobPilot is built so the expensive LLM only ever sees a handful of finalists.
Layer A — pure Python, ZERO LLM. All scraping, deduplication, and hard filtering run as
plain python3 scripts via bash. These scripts MUST NEVER call the LLM. They read config and
the SQLite cache, hit REST APIs, and write JSON to /tmp. A run that scrapes hundreds of jobs
narrows to ~5–15 survivors here at no token cost.
Layer B — LLM reasoning. Only the Layer-A survivors reach the LLM: semantic ATS scoring,
salary research, resume tailoring, and the final notification. This keeps a daily run cheap
enough to live inside a Pro subscription.
Slash commands
| Command | What it does |
|---|---|
/job-setup |
One-time wizard. Collects preferences + parses your resume into a profile. |
/job-search |
The main pipeline. Scrape -> dedupe -> filter (Layer A) -> score -> research -> report -> tailor -> notify (Layer B). |
/job-tailor <job_id|URL|JD text> |
Tailor your resume to a single job and ATS-score before/after. |
/jobpilot-clear |
Wipe cached job IDs, score cache, and reports. Keeps preferences + resume. |
resume-validate |
Internal skill used by /job-search to score one JD against your profile. Not called directly. |
Where things live
- Repo (this plugin):
~/projects/jobpilot - User data (created by
setup.sh, NOT in the repo):~/.claude/job-hunt-ai/options/preferences.json— your search criteriacache/jobs.sqlite— seen-jobs + score cachecache/profile.json— parsed resumeresumes/—base.tex/base.docxandtailored/outputsreports/— generated CSVs.env— secrets
Rule for every Layer-A script
Scripts under scripts/ named apify_scraper.py, dedupe.py, and filter.py run via bash
and must never invoke the LLM. They only read config, call REST APIs, and read/write the cache
and /tmp JSON files. Secrets are always loaded through scripts/secrets.py, never read from
env vars directly.