apply-job

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: apply-job
description: Automated job applications with AI-powered resume tailoring, cover letters, recruiter outreach, interview prep, ATS optimization, and smart analytics.
license: MIT
metadata:
  author: theaayushstha1
  version: "4.0.0"
allowed-tools:
  - Read
  - Write
  - Edit
  - Glob
  - Grep
  - WebFetch
  - WebSearch
  - mcp__playwright__browser_navigate
  - mcp__playwright__browser_snapshot
  - mcp__playwright__browser_click
  - mcp__playwright__browser_type
  - mcp__playwright__browser_fill_form
  - mcp__playwright__browser_file_upload
  - mcp__playwright__browser_take_screenshot
  - mcp__playwright__browser_wait_for
  - mcp__playwright__browser_press_key
  - mcp__playwright__browser_select_option
---

# Job Application Agent v4

Automated job application system with smart scoring, ATS optimization, interview prep, recruiter outreach, and analytics.

## Commands

```bash
/apply-job <url>                 # Full application workflow
/apply-job search <query>        # Search + batch apply with smart filtering
/apply-job outreach <url>        # Recruiter outreach only
/apply-job status                # View statistics
/apply-job status weekly         # Weekly analytics digest
/apply-job score <url>           # Score job fit only
/apply-job prep <company>        # Interview prep mode
/apply-job followup              # Check follow-up reminders
/apply-job ats-check <url>       # ATS keyword analysis
/apply-job dedup                 # Find duplicate applications
```

## Setup Required

| File | Purpose |
|------|---------|
| `$APPLIER_HOME/data/profile.json` | Your personal info (from template) |
| `$APPLIER_HOME/data/applications.json` | Application tracking |
| `$APPLIER_HOME/resume/master_resume.md` | Your complete resume |
| `$APPLIER_HOME/templates/cover_letter.md` | Cover letter template |
| `$APPLIER_HOME/templates/cold_email.md` | Cold email template |

**$APPLIER_HOME** = `~/job-applier-agent` (cross-platform)

See [SETUP.md](SETUP.md) for installation instructions.

---

## Security and Privacy

### Credential Handling
- No passwords or tokens are stored by this skill
- Browser actions use the user's existing authenticated sessions (Gmail, LinkedIn)
- All personal data stays in local files (`profile.json`, `applications.json`)
- No data is sent to external servers beyond the user's own accounts

### Input Sanitization
- Job descriptions fetched via Playwright are treated as untrusted input
- JD content is only used for text comparison (keyword matching, scoring)
- JD content is never executed as code or injected into scripts
- Resume tailoring modifies only markdown content, no code generation or execution

### User Confirmation Required For
- Submitting applications (pre-submit screenshot shown)
- Sending emails via Gmail
- Sending LinkedIn connection requests
- Any action when fit score < 5.0 or dealbreakers detected

### Rate Limiting
- LinkedIn: max 10-15 connections per session
- Emails: sent one at a time with user's own Gmail
- Applications: sequential, not parallelized

---

## Core Principles

### Workflow Automation
- Proceed through routine steps (form filling, screenshot capture) without pausing
- Always pause and confirm before: submitting applications, sending emails, sending LinkedIn requests
- Stop immediately for: login walls, CAPTCHA, critical errors, fit score < 5.0, dealbreakers detected

### Outreach Order
```
1. Find recruiters on LinkedIn (names, titles, URLs)
2. Find publicly available contact info
3. Send cold emails via Gmail FIRST (confirm before sending)
4. THEN send LinkedIn connection requests (confirm before sending)
5. Track everything in applications.json
```
**Email = PRIMARY (higher response rate). LinkedIn = SECONDARY.**

---

## Workflow: Apply to Job URL

### Step 1: Dedup Check
```
1. Read $APPLIER_HOME/data/applications.json
2. Check if URL already exists in applications[]
3. Check if same company + similar role title exists (fuzzy match)
4. If duplicate found: