name: evaluate-job
description: Evaluate LinkedIn job postings for fit. This skill should be used when the user shares a LinkedIn job URL and wants analysis of whether to apply. Triggers on LinkedIn job links, "evaluate this role", or requests to analyze job fit.
requires: browser-automation
platform_note: Requires browser automation with LinkedIn login. Works on Claude Code (Chrome MCP) and OpenClaw (browser tool). OpenCode needs external browser or tavily-extract fallback.

Evaluate Job

Analyze LinkedIn job postings against user's background, current pipeline health, and career criteria. Output a recommendation (APPLY/CONSIDER/PASS) with structured reasoning.

Workflow

1. Navigate to job posting:
2. OpenClaw: Use browser tool → snapshot or screenshot to read the page
3. Claude Code: Use Chrome extension (profile="chrome") or web_fetch
4. Fallback: mcporter call tavily.tavily-extract urls='["URL"]'
5. Extract: company name, role title, requirements, responsibilities, preferred qualifications, salary if disclosed, applicant stats

Why browser over fetch for LinkedIn:
- Fetch/extract misses applicant stats (count, seniority breakdown) — requires being logged in
- Browser sees the full page as logged-in user, including competition metrics
- Can click "Show more" to expand full JD if needed

Extraction tip: read_page captures the full accessibility tree including content below the viewport — no scrolling needed. Just navigate, wait 2 seconds for load, then read_page once to get the complete JD.

1. Check for duplicates (before full analysis):
2. Search vault for existing note matching company + role (e.g., [[*Role* - *Company*]])
3. Check [[Job Hunting]] "Applied Jobs" and "Passed On" sections for the company name
4. If match found:
   • Show user what was found (existing note, application status, date)
   • Ask: "Already evaluated/applied to [match]. Proceed with analysis anyway?"
   • If user says no, stop early — no further analysis needed

5. This prevents duplicate work and catches related roles at the same company

6. Load context files (can run in parallel with step 1) — Check user's CLAUDE.md for background, credentials, differentiators, current situation, and job hunting status

7. Analyze fit across dimensions (see Fit Dimensions below)

8. Factor pipeline health:

9. Healthy (5+ active, interviews scheduled): Maintain standards — PASS on poor fits
10. Thin (<5 active, no interviews): Lower bar — CONSIDER "good enough" roles

11. Consider urgency of user's situation when weighing trade-offs

12. Check for red flags (feedback loop from /debrief):

13. Review [[Job Hunting]] → Market Signals for relevant patterns:
    • Objections: What got pushback in similar roles?
    • Wins: What hooks landed that this role could use?
    • Persona Mismatches: Did similar roles expect different positioning?
14. Review [[Job Hunting]] → Anti-Signals for known rejection patterns

15. If role matches a pattern, flag it in analysis and factor into recommendation

16. Output recommendation: APPLY, CONSIDER, or PASS with clear reasoning

17. Review with Judge:

18. Run evaluation through /judge with job-eval criteria
19. Check: fit_analysis, red_flags, specificity, recommendation, career_direction
20. If verdict is needs_work: revise analysis (max 2 iterations)

21. Ensures recommendation is specific and actionable, not vague

22. Create vault note — MANDATORY for ALL outcomes:

23. Do this immediately after giving recommendation — don't wait for user to ask
24. Filename: [[Role Title - Company]]
25. MUST include full JD details: Copy requirements, responsibilities, qualifications verbatim from the posting
26. Include: Fit analysis table, recommendation reasoning

27. This creates a record for future reference, pattern recognition, and duplicate detection

28. Update job tracking:

    • APPLY: Add to "Applied Jobs" or "To Apply" section in [[Job Hunting]]
    • PASS: Add to "Passed On" section with one-line reason
    • CONSIDER: Note in appropriate section with context

29. If APPLY:

    • Easy Apply roles: Ask whether to proceed with application now
    • Company website roles: Add to "To Apply" list (external ATS requires more effort; note for later)

Note: No reliable way to close browser tabs via MCP — leave tab open for user to close manually.

Fit Dimensions

Note Template

# [Role Title], [Company]

**Source:** [LinkedIn URL]
**Discovered:** [Date]
**Status:** [Applied/Passed/Considering]

## Job Details

**Title:**
**Company:**
**Location:**
**Salary:** [if disclosed]
**Applicants:** [count] ([seniority breakdown])

## Requirements

[COPY FULL REQUIREMENTS FROM JD - verbatim bullet points]

## Responsibilities

[COPY FULL RESPONSIBILITIES FROM JD - verbatim bullet points]

## Preferred/Nice-to-Have

[If listed separately in JD]

## Fit Analysis

| Dimension | Assessment | Notes |
|-----------|------------|-------|
| Seniority | ✅/⚠️/❌ | |
| Role Focus | ✅/⚠️/❌ | |
| Industry | ✅/⚠️/❌ | |
| Tech Stack | ✅/⚠️/❌ | |
| Competition | ✅/⚠️/❌ | |
| Salary | ✅/⚠️/❌ | |
| Anti-Signal | ✅/⚠️/❌ | |

## Recommendation

**[APPLY/CONSIDER/PASS]**

[Reasoning]

Passed On Format

- [[Role Title - Company]] (Date) - [One-line reason]

Batch Processing

For multiple jobs, run this skill sequentially on each URL. Start with quick duplicate check before full analysis.

Related Skills

• chrome-automation — Browser best practices (read_page, wait times, tab creation)
• vault-pathfinding — Vault paths and conventions (where to read/write notes)
• /review-saved-jobs — Batch processing of LinkedIn saved jobs (chains to this skill)
• /debrief — Captures interview signals that feed back into this skill's red flag check