job-matcher

name	job-matcher
description	Use when users ask to find jobs, match jobs against a resume, search Boss直聘 or Google Jobs, rank job listings, or need help with "找工作", "职位匹配", "推荐职位", "job search", or "find jobs for me".

Job Matcher

AI-powered job matching that searches live listings and scores them against your resume. Supports Boss直聘 (via OpenCLI) and Google Jobs (via SerpApi).

When to Use

Use this skill when the user:

• Wants to find jobs matching their resume or background
• Asks to search Boss直聘, Google Jobs, or job platforms in general
• Has a resume and wants ranked job recommendations
• Says "找工作", "职位匹配", "推荐职位", "find jobs", "match jobs", "job search"
• Wants to compare job listings against their skills and preferences

Do not use this skill when:

• The user wants career advice without job searching
• The user wants interview prep or mock interviews
• The user wants resume editing or formatting help

Prerequisites

For Boss直聘 (Chinese job market)

1. OpenCLI: npm install -g @jackwener/opencli
2. Browser Bridge: Install the Chrome extension (see https://github.com/jackwener/opencli)
3. Boss直聘 login: Open Chrome, go to https://www.zhipin.com, log in
4. Verify: Run opencli doctor — should show Browser Bridge connected

For Google Jobs (international)

1. SerpApi key: Sign up at https://serpapi.com/
2. Set environment variable: export SERPAPI_API_KEY=your_key_here
3. Verify: curl "https://serpapi.com/search.json?engine=google_jobs&q=test&api_key=$SERPAPI_API_KEY" should return results

If neither source is available, inform the user and help them set one up.

Workflow

Phase 1: Resume Input and Profile Extraction

Accept the user's resume via:

• File path: Read MD or TXT directly. For PDF files, extract text using one of these methods (try in order):
  1. python3 -c "from pypdf import PdfReader; print('\n'.join(p.extract_text() or '' for p in PdfReader('FILE').pages))" (if pypdf installed)
  2. pdftotext FILE - (if poppler-utils installed)
  3. strings FILE | head -500 (last resort — noisy but works everywhere)
• Pasted text: User pastes resume content directly

Extract a structured profile:

{
  "target_roles": ["Backend Engineer", "Go Developer"],
  "skills": ["Go", "Python", "Kubernetes", "MySQL", "Redis"],
  "experience_years": 5,
  "experience_level": "3-5年",
  "education": "本科",
  "preferred_cities": ["杭州", "上海"],
  "preferred_industries": ["互联网", "人工智能"],
  "salary_expectation": "30-50K",
  "dealbreakers": ["996", "外包"],
  "strengths": ["Distributed systems", "High concurrency"],
  "career_highlights": ["Built microservice architecture from scratch"]
}

Show the extracted profile and ask for confirmation before proceeding.

For field options and valid values, see references/config-reference.md.

Phase 2: Job Fetching

Boss直聘 (OpenCLI)

Run multiple searches to cover different angles:

# Primary — exact role
opencli boss search "Go后端工程师" --city 杭州 --experience 3-5年 --salary 30-50K --limit 30 -f json

# Secondary — broader terms
opencli boss search "后端开发" --city 杭州 --experience 3-5年 --salary 30-50K --limit 30 -f json

# Tertiary — different city
opencli boss search "Go后端工程师" --city 上海 --experience 3-5年 --salary 30-50K --limit 30 -f json

Parameter mapping:

Profile Field	OpenCLI Flag	Example Values
experience_level	--experience	应届/1年以内/1-3年/3-5年/5-10年/10年以上
education	--degree	大专/本科/硕士/博士
salary_expectation	--salary	3K以下/3-5K/5-10K/10-15K/15-20K/20-30K/30-50K/50K以上
preferred_cities	--city	北京/上海/广州/深圳/杭州/成都/南京/武汉/西安 etc.
preferred_industries	--industry	互联网/电子商务/游戏/人工智能/大数据/金融 etc.

Deduplicate results by security_id. Expect 30-90 unique jobs across searches.

Google Jobs (SerpApi)

curl "https://serpapi.com/search.json?engine=google_jobs&q=Backend+Engineer&location=San+Francisco%2C+California%2C+United+States&api_key=$SERPAPI_API_KEY"

Key parameters:

Parameter	Description	Example
q	Search query (role + skills)	Backend Engineer Go Kubernetes
location	Full location string	San Francisco, California, United States
chips	Filter chips	date_posted:week for recent jobs

SerpApi requires full location format. See references/config-reference.md for the location mapping table.

Run 2-3 searches with different query variations to maximize coverage.

Phase 3: Quick Filter (Rule-Based)

Apply hard filters to eliminate mismatches:

1. Salary mismatch: Job salary doesn't overlap with expectation
2. Experience mismatch: Job requires significantly different experience level
3. Dealbreaker match: Company name or title matches dealbreakers
4. Duplicate company: Keep only the best-matching role per company

Report filter stats:

📊 Quick filter results:
- Total jobs fetched: 72
- Salary mismatch: -18
- Experience mismatch: -8
- Dealbreaker match: -5
- Entering deep evaluation: 41 jobs

Phase 4: Deep Evaluation (LLM)

For remaining candidates (typically 20-40), fetch full job descriptions:

Boss直聘:

opencli boss detail <security_id> -f json

Fetch in batches of 5 with 2-3 second delays between batches to avoid rate limiting.

Google Jobs: Full descriptions are already included in SerpApi results.

Score each job on these dimensions:

Dimension	Weight	Criteria
Skills Match	30%	How many required skills does the user have? Core vs nice-to-have?
Experience Match	20%	Does experience level and domain match?
Salary Match	15%	Is salary within or above expectation?
Role Level	15%	Does role level match career stage?
Company/Industry	10%	Does company industry, scale, stage match preferences?
JD Quality	10%	Is JD detailed? Is the recruiter active? Red flags?

For each job, generate:

• Score: 0-100
• Match reasons: 2-3 bullet points
• Concerns: Potential mismatches or red flags
• Recruiter activity (Boss直聘 only): From active_time field

Phase 5: Present Results

Sort by score descending. Present the top 10:

## 🎯 Top 10 Job Recommendations

| # | Score | Position | Company | Salary | Key Match |
|---|-------|----------|---------|--------|-----------|
| 1 | 95 | Senior Go Engineer | ByteDance | 40-60K | Full skill match, distributed systems |
| 2 | 88 | Backend Tech Lead | XX Tech | 35-50K | Tech stack match, management exp ✅ |
| 3 | 82 | Go Developer | YY Corp | 30-45K | Core skills match, smaller company |

📊 Pipeline: 72 fetched → 41 after filter → Top 10 ranked

For each recommended job, provide a detail card:

### #1 Senior Go Engineer — ByteDance (95/100)

Salary: 40-60K·15months | Experience: 3-5 years | Education: Bachelor+
Location: Hangzhou · Yuhang · Future Sci-Tech City
Skills: Go, Kubernetes, MySQL, Redis, Microservices
Company: Internet | Series D+ | 10000+

✅ Match:
- Go + distributed systems experience is a direct fit
- High concurrency experience is a bonus
- Salary range covers expectation

⚠️ Watch out:
- JD mentions "moderate overtime expected"

🔗 https://www.zhipin.com/job_detail/xxx.html

Phase 6: User Interaction

After presenting results, offer these actions:

1. View details: "Tell me more about #3"
2. Adjust filters: "Widen salary to 20-50K and re-search"
3. Open links: "Open #1, #3, #5 in browser"
4. Export: "Export recommendations to a file"

Opening links:

open "https://www.zhipin.com/job_detail/xxx.html"
# or for Google Jobs links
open "https://..."

Exporting: Write results to ~/job-matches-YYYY-MM-DD.md.

IMPORTANT — No auto-greet: OpenCLI's boss greet and boss send are recruiter-side commands (Boss identity), not job-seeker commands. Never auto-apply or auto-greet. Always direct users to open links and apply manually.

Language Rule

• Detect the language of the user's first message
• Chinese input → respond in Chinese
• English input → respond in English
• Maintain the same language throughout the session

Error Handling

Error	Action
opencli doctor fails	Guide user through OpenCLI + Browser Bridge setup
opencli boss search returns empty	Try broader terms, different city, or remove salary/experience filters
opencli boss detail fails	Skip job, note "Position may have been taken down"
SerpApi returns error	Check API key, quota, and location format
Rate limiting (empty responses)	Add longer delays, reduce batch size
Chrome not logged in	Tell user to open Chrome → zhipin.com → log in
Boss直聘 IP restriction	Warn user that Boss直聘 may block requests from certain IPs/regions. Suggest switching to Google Jobs or using a Chinese IP

Common Mistakes

• Searching without extracting a profile first — always understand the resume before searching
• Using only one search query — run 2-3 variations for broader coverage
• Ignoring rate limits — Boss直聘 blocks rapid requests, always add delays
• Fabricating job data — only present what the search tools actually return
• Auto-greeting or auto-applying — the greet/send commands are recruiter-side only
• Using short location names with SerpApi — always use full format (e.g., "San Francisco, California, United States")
• Not deduplicating results across multiple searches

Bundled Reference

Read references/config-reference.md for valid field values: cities, salary ranges, experience levels, education levels, dealbreakers, and SerpApi location mappings.