job-hunter

name	job-hunter
description	Primary job search command center. Routes to individual skills for all job search work.
when_to_use	Use for all job search work: running hunts, processing candidates, tailoring, researching, and managing the pipeline.
argument-hint	[batch|one <url>|finalize|tailor <job>|outreach <job>|interview <job>|score <job>|research <co>|stories|linkedin <cmd>|add-region|style|setup|doctor|dashboard|help]
disable-model-invocation	true
allowed-tools	Bash Read Edit Write WebSearch WebFetch
author	Abdul Basit (@abdulrbasit)
category	workflow

Execute .claude/skills/caveman/SKILL.md inline before processing any command. Execute .claude/skills/job-hunter/_rules.md inline before processing any command.

Job Hunter Command Center

One entry point for all job search work. Keep output compact — no raw candidate snapshots, full logs, PDFs, or Draft/raw story-bank sections.

Arguments: $ARGUMENTS

Key Paths

What	Path
Story bank	profile/story_bank.md
Candidate queue	outputs/state/agent_candidate_queue.json
Job outputs	outputs/jobs/<slug>/
Job DB	outputs/state/jobs.db
Config	config/job_hunter.yml

Safety Rules

• Never fabricate resume facts, employers, dates, metrics, skills, or job history.
• Never submit applications, send messages, connect, follow, like, comment, or post automatically.
• Never overwrite profile/config inputs unless that mode requires it and the user requested it.
• Leave generated changes uncommitted unless the user runs job-hunter finalize.

Initialization

Before any mode that reads profile files (batch, one, tailor, score, screen, interview, outreach), run:

job-hunter internal compile-profile

This compiles profile files into minified versions for the session. A missing config or profile file is a silent no-op (fine for a fresh workspace). Any other failure logs one compact warning line to job_hunt.log and falls back to the raw profile files — it never raises or blocks the session.

Routing

Normalize the first argument to lowercase. Empty argument → help.

Daily workflow

• dashboard, apps, applications: tell the user to open the Job Hunter app (job-hunter dash) and use the Applications view — there is no terminal dashboard to run inline.
• batch, process, queue: execute .claude/skills/job-hunter/modes/batch.md inline.
• one <url>, url <url>, or any pasted http(s):// URL: execute .claude/skills/job-hunter/modes/one.md inline with the URL and remaining arguments.
• finalize: run job-hunter finalize (add --push if the user asked to push). It validates, commits, and optionally pushes durable outputs itself — no separate skill needed. Show its output verbatim.
• screen: execute .claude/skills/job-hunter/modes/screen.md inline.

Per-job actions — second token is the job slug

• tailor <job>: execute .claude/skills/job-hunter/modes/tailor.md inline.
• outreach <job>: execute .claude/skills/job-hunter/modes/outreach.md inline.
• interview <job>: execute .claude/skills/job-hunter/modes/interview.md inline.
• score <job>: execute .claude/skills/job-hunter/modes/score.md inline.
• research <job>: execute .claude/skills/job-hunter/modes/research.md inline.
• stories: execute .claude/skills/job-hunter/modes/stories.md inline.

LinkedIn sub-router — linkedin (alone or with sub-argument): execute .claude/skills/linkedin/SKILL.md inline with remaining arguments.

Setup sub-router — setup, init, onboard, doctor, health, check, add-region, region, style: execute .claude/skills/setup/SKILL.md inline with remaining arguments.

Unknown mode → print the command menu and ask the user to choose a listed mode.

Command Menu

Job Hunter Command Center

── Daily Workflow ──────────────────────────────────────────────────────
/job-hunter dashboard          Open the Job Hunter app to review applications
/job-hunter batch              Process the next frozen candidate batch
/job-hunter one <url>          Process one job URL end-to-end
/job-hunter finalize           Runs `job-hunter finalize` to commit durable reviewed outputs
/job-hunter screen             Pre-screen a frozen batch against config exclusion rules

── Per-Job Actions ─────────────────────────────────────────────────────
/job-hunter tailor <job>       Tailor resume + cover letter for a job
/job-hunter outreach <job>     Draft LinkedIn connection + follow-up
/job-hunter interview <job>    Generate predicted interview questions
/job-hunter score <job>        Score one job 0-100 vs. resume and story bank
/job-hunter research <job>     Web-search a company for an imported job
/job-hunter stories            Refine raw work notes into rated STAR stories

── LinkedIn ────────────────────────────────────────────────────────────
/job-hunter linkedin ideas     Generate weekly LinkedIn post ideas
/job-hunter linkedin draft     Write one ready-to-post LinkedIn draft
/job-hunter linkedin engage    Draft comments for posts in your feed
/job-hunter linkedin network   Build a weekly connection queue

── Tools ───────────────────────────────────────────────────────────────
/job-hunter setup              One-time onboarding for a fresh workspace
/job-hunter doctor             Run the health checker and show setup status
/job-hunter add-region         Add/remove a region — dash → Settings → Guided → Regions
/job-hunter style              Change resume color/font — dash → Settings → Resume Style

Output Rules

• Execute child skills inline from their SKILL.md; do not print a slash command as a handoff.
• Print only paths, counts, short decisions, and next actions.
• Leave generated changes uncommitted unless the user runs job-hunter finalize.