AI Job Application Agent - Claude Code Instructions

Overview

This is a job application automation toolkit. It automates filling and submitting applications on LinkedIn Easy Apply, Greenhouse, Lever, Jobvite, and Ashby, plus email triage and Google Sheets tracking.

Two ways to drive it:

1. Bundled skills (recommended) — type /job-coach, /job-apply, /job-evaluate, /job-cv, /job-track, /job-triage, /job-status, /job-outreach, /job-followup, /job-dashboard, /job-interview, /job-patterns, or /job-setup in any Claude Code session. Each skill wraps the scripts below and renders results as a markdown table. Install once with bash skills/install.sh.
2. Raw scripts — call the Node.js / Python scripts in scripts/ directly from your terminal (documented below).

Bundled Skills

Installed by bash skills/install.sh (symlinks them into ~/.claude/skills/). Each skill's prompt lives at skills/<name>/SKILL.md — read it to see exactly what the agent does.

First-run flow (gstack-style): open Claude Code anywhere and paste this one message:

Install ai-job-agent: run git clone --single-branch --depth 1 https://github.com/AkbarDevop/ai-job-agent ~/.claude/skills/ai-job-agent && cd ~/.claude/skills/ai-job-agent && bash skills/install.sh then add an "ai-job-agent" section to my CLAUDE.md telling you to treat me like a career-coaching client — open-ended job-search talk → /job-coach (intake + research + slate), job URL alone → /job-apply, "evaluate this URL" → /job-evaluate (rubric + report + tailored CV), "tailor my CV" → /job-cv, "prep me for X interview" → /job-interview, "what's working / why am I getting rejected" → /job-patterns, reaching out to a person → /job-outreach, follow-ups → /job-followup, "how am I doing" → /job-dashboard, tracker → /job-track, status updates → /job-status, Outlook → /job-triage, first-time setup → /job-setup. Never answer job-search questions conversationally — always dispatch. Tell me when done, then run /job-setup and chain into /job-coach intake.

Claude clones into ~/.claude/skills/ai-job-agent/, registers all 13 skills as symlinks under ~/.claude/skills/, writes the coach-first routing block into CLAUDE.md, runs /job-setup (identity + resume + optional msmtp), and chains into /job-coach intake (target roles + companies + timeline → live market research → ranked slate of next moves). After that, the user talks naturally — "I want to apply to this url", "email the VP at X", "evaluate this posting", "tailor my CV for this JD", "prep me for my interview at Y", "what's working" → diagnostics, "how am I doing" — and Claude dispatches to the right skill. No cd required ever; skills work from any directory in any session. bash wizard.sh and bash setup.sh still exist for non-Claude-Code users.

Skills find this repo via $AI_JOB_AGENT_ROOT → ~/.claude/skills/ai-job-agent/REPO_PATH → ~/ai-job-agent. Set the env var or rerun install.sh from a non-default clone location.

Unified mode — npm run agent

Daily driver. Opens Claude Code chat (top pane) + live TUI dashboard (bottom pane) in one terminal window via tmux. They auto-sync via fs.watch — anything Claude does in chat that touches application-tracker.csv or outreach-log.csv re-renders the dashboard within ~200ms.

Launcher: bin/job-agent.sh. If tmux isn't installed it offers to brew install tmux (macOS) or apt install tmux (Linux) with a one-key confirmation. On decline, falls back to opening two Terminal/iTerm tabs (macOS) or printing manual instructions (Linux).

NPM scripts

Output locations (all gitignored)

Configuration

All personal details are in the config/ directory:

• config/candidate-profile.md -- Full candidate profile, application rules, session state, and agent handoff context. Read this first in every session.
• config/answer-bank.md -- Pre-written answers for common application questions.
• config/linkedin-config.json -- LinkedIn Easy Apply config with identity, resume path, and Chrome cookie path.
• config/example-config.json -- Reference example showing all available fields.

Workflow

Recommended (skill-driven): /job-coach (intake + slate) → for each row: /job-evaluate <url> (deep-dive + rubric) → if fit ≥ 4.0, automatically chains into /job-cv (tailored PDF) and offers /job-apply → after applying, run /job-followup weekly → /job-triage to log replies → /job-status to flip rows → /job-patterns once a month for diagnostics → /job-interview whenever an interview shows up.

Legacy (raw scripts): Search → Apply → Track → Triage.

1. Search for jobs

Use /job-search or python-jobspy to find matching roles. Filter by:

• Role fit (match resume variants to role types)
• Work authorization eligibility
• Location preferences
• Recency (prefer jobs posted in the last 72 hours)

2. Apply to jobs

Choose the right script based on the ATS platform:

# LinkedIn Easy Apply
node scripts/linkedin-easy-apply.js <jobUrl> config/linkedin-config.json

# Lever
node scripts/lever-apply.js <jobUrl> <configPath>

# Greenhouse
node scripts/greenhouse-apply.js <jobUrl> <configPath>

# Jobvite
node scripts/jobvite-apply.js <jobUrl> <configPath>

# Ashby
node scripts/ashby-apply.js <jobUrl> <configPath>

For Lever, Greenhouse, Jobvite, and Ashby, you need to inspect the form first to build the config with the correct field IDs and values.

3. Track applications

After each successful submission:

1. Append a row to the local application-tracker.csv
2. Append an entry to the daily log submitted-applications-YYYY-MM-DD.md
3. Sync to Google Sheets: python3 scripts/google-sheet-sync.py application-tracker.csv

4. Triage emails

Use the Outlook scripts to search for confirmation and rejection emails:

# Search for job-related emails
node scripts/outlook-triage.js search "application confirmation"

# Read a specific email
node scripts/outlook-triage.js extract 0

# Mark as read after processing
node scripts/outlook-triage.js mark-read 0

5. Update statuses

When you receive rejections, interviews, or other updates:

python3 scripts/tracker-status-update.py updates.json

Application Rules

1. Truthfulness first: Prefer truthful, submittable applications over aggressive volume.
2. Work authorization: Skip hard citizen/green-card gated roles unless the form provides a truthful path for the candidate's actual status.
3. Resume selection: Use the appropriate resume variant for each role type (see candidate-profile.md for the mapping).
4. Captcha honesty: If a form stalls behind CAPTCHA, log it as "blocked" -- never inflate submitted counts.
5. Headless default: Use headless browser by default. Only open visible windows when manual intervention (e.g., CAPTCHA solving) is required.
6. Email triage: Mark confirmation emails as read after logging. Leave interviews, assessments, and recruiter outreach unread.

Available Scripts

Exit Codes

All application scripts use consistent exit codes:

• 0 -- Success (submitted or ready in dry-run mode)
• 1 -- Crash or unexpected error
• 2 -- Blocked on unknown required field
• 3 -- Captcha or submission timeout
• 4 -- Step limit exceeded (LinkedIn only)

Environment Variables

Recommended Companion Skills

The skills below are community-built and complement the 13 bundled skills above:

• /job-search -- Multi-board job search
• /tailor-resume -- Resume customization per posting
• /apply -- AI-assisted ATS form filling (different from bundled /job-apply, which wraps this repo's scripts directly)
• /interview-prep-generator -- STAR stories and practice questions
• /resume-ats-optimizer -- ATS keyword optimization
• /salary-negotiation-prep -- Market rate research

See skills/README.md for the full list and install commands.

gstack

This repo pairs well with gstack — a general-purpose Claude Code skill pack for shipping code. Install gstack once (git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup) and these gstack skills become useful in this repo's workflow:

• Web browsing: use /browse for all web browsing. Never use mcp__claude-in-chrome__* tools. /connect-chrome + /setup-browser-cookies for authenticated pages (useful for LinkedIn / Outlook flows).
• Shipping code: /ship to sync main + run tests + push + open PR (this repo pushes direct to main by convention, but /ship still handles the rest). /review for pre-landing paranoid review. /codex for a second-opinion code review.
• Testing: /qa / /qa-only for systematic QA + fix loops. /browse for one-off dogfooding.
• Debugging: /investigate for root-cause debugging. /careful / /freeze / /guard / /unfreeze for safety during destructive ops on the trackers / config files.
• Dev-experience: /devex-review after shipping a new skill — it actually runs the install + first-skill-invocation and scores the onboarding. /plan-devex-review before adding new skills.
• Post-ship: /document-release to sync README.md / CLAUDE.md / skills/README.md after a skill lands. /retro for the weekly "what did we ship" log.

If gstack isn't installed, those skills are optional — this repo's 13 bundled skills (/job-coach, /job-setup, /job-evaluate, /job-apply, /job-track, /job-triage, /job-status, /job-outreach, /job-followup, /job-dashboard, /job-cv, /job-interview, /job-patterns) work standalone.

File Locations

• Config templates: config/*.template.*
• Active configs: config/linkedin-config.json, config/candidate-profile.md, config/answer-bank.md
• Scripts: scripts/
• Templates: templates/
• Documentation: docs/
• Local tracker: application-tracker.csv (created by setup.sh)
• Daily logs: submitted-applications-YYYY-MM-DD.md (created per session)