JobClaw 🦀

AI-powered job search automation as an OpenClaw skill. Search LinkedIn and Indeed every day. Score results against your profile. Track everything in a local CSV. Zero backend, zero dashboard — just you and your data.

---

What It Does

• Multi-platform search: LinkedIn + Indeed via python-jobspy
• Two search modes: coding (ML engineer, data scientist, AI engineer) and noncoding (research, no-leetcode, consulting roles)
• AI scoring: Jobs scored 0–100 against your skill profile. Only qualified jobs (≥70 by default) are saved.
• CSV tracker: All jobs saved to jobs.csv with 21 columns — no database server needed
• Smart dedup: Won't add the same company+role or URL twice
• Daily automation: One cron job, one notification each morning
• /newjob <url>: Paste any job URL and get an instant score + analysis

---

Requirements

• Python 3.10+
• OpenClaw (for skill integration and notifications)
• python-jobspy:

  pip install python-jobspy

---

Installation

Via ClawHub (recommended)

clawhub install jobclaw

Manual

git clone https://github.com/weixijia/jobclaw ~/.npm-global/lib/node_modules/openclaw/skills/jobclaw

---

Contributing

Pull requests welcome. The CI pipeline auto-publishes to ClawHub on every push to main — no manual release step needed.

---

First Run

Talk to your OpenClaw agent:

"Set up my job search"

The agent walks you through everything conversationally — one question at a time:

1. Your name
2. Background — a sentence about what you do (used for job scoring context)
3. Key skills — e.g. python, pytorch, sql, docker
4. Target roles — e.g. ML engineer, data scientist
5. Locations — e.g. London, UK, Remote, Berlin, DE
6. Search preferences — how recent, minimum score (defaults work fine)
7. Daily schedule — what time to run, timezone
8. Notifications — Telegram bot, OpenClaw channel, or none

At the end, the agent writes ~/Documents/JobClaw/config.json and offers to run the first search immediately. No editing files, no CLI wizard.

---

Example Config

{
  "user": {
    "name": "Alex",
    "background": "Software engineer with ML experience",
    "skill_keywords": ["python", "pytorch", "machine learning", "deep learning"]
  },
  "search": {
    "locations": ["London, UK", "Remote"],
    "min_score": 70,
    "hours_old": 48,
    "platforms": ["linkedin", "indeed"]
  },
  "notifications": {
    "enabled": true,
    "telegram_bot_token": "your_bot_token",
    "telegram_chat_id": "your_chat_id"
  },
  "schedule": {
    "daily_time": "07:30",
    "weekdays_only": true,
    "timezone": "Europe/London"
  }
}

Another example — data science background:

{
  "user": {
    "name": "Sam",
    "background": "Data scientist, fintech experience",
    "skill_keywords": ["pandas", "scikit-learn", "sql", "statistics", "forecasting"]
  },
  "search": {
    "locations": ["New York, NY", "San Francisco, CA", "Remote"],
    "min_score": 65
  }
}

See jobclaw/assets/config.example.json for the full template with all options.

---

Usage

Chat commands (inside OpenClaw)

Command	What happens
run job search	Full search (coding + noncoding)
run coding job search	ML engineer, AI engineer, data scientist roles
run noncoding job search	Research, consulting, no-leetcode roles
/newjob https://...	Analyse a specific job posting
show job stats	Total, by status, new today
top jobs	Top 10 by match score
search for "LLM engineer" in Berlin	Custom keyword + location
set up daily search at 8am	Configure OpenClaw cron

CLI commands

# Full search
python3 scripts/search.py --mode all

# Coding roles only
python3 scripts/search.py --mode coding

# Research / no-leetcode roles only
python3 scripts/search.py --mode noncoding

# Preview without saving
python3 scripts/search.py --mode all --dry-run

# Custom keyword and location
python3 scripts/search.py --mode coding --keyword "MLOps engineer" --location "Berlin, DE"

# Tracker stats
python3 scripts/tracker.py stats
python3 scripts/tracker.py top

---

How Scoring Works

1. Base score: 50
2. Each matching positive keyword in the job title/description: +5 (capped at 100)
3. Negative keywords (e.g. "intern", "unpaid"): −15 each
4. Keywords in your skill_keywords config are added to the positive list

Coding mode positive keywords: machine learning, deep learning, pytorch, llm, nlp, computer vision, mlops, rlhf, fine-tuning, generative ai …

Noncoding mode positive keywords: healthcare, clinical, research scientist, fintech, fraud, postdoc, product manager, consulting …

Adjust min_score in config (default: 70) for more or fewer results.

---

Interview Type Classification

JobClaw tags each job with an estimated interview type:

Type	Description	Typical for
Research Talk	Presentation + Q&A, no coding	Postdoc, Research Scientist at labs
No Leetcode	Behavioural + ML knowledge	Startups, many mid-size companies
Case Study	Business/ML case + presentation	Consulting, pharma, finance
Take-Home	1–3 day ML task	Startups
Fair Coding	Basic Python + ML discussion	Applied scientist roles
Standard Coding	LeetCode medium	Most tech companies
Heavy Leetcode	LeetCode hard	Big tech (FAANG)

---

CSV Schema

~/Documents/JobClaw/data/jobs.csv:

Column	Description
company	Company name
role	Job title
location	City, country
work_mode	Remote / Hybrid / On-site / Unknown
salary	Salary if listed
job_url	Link to posting
date_posted	When posted (YYYY-MM-DD)
date_found	When JobClaw found it
source	Daily Search / Manual
match_score	0–100 relevance score
ml_direction	e.g. NLP/LLM, Healthcare AI, General ML/AI
seniority	Junior / Mid / Senior / Lead
interview_type	Research Talk / No Leetcode / Standard Coding / …
status	New / Applied / Phone Screen / Offer / Rejected / …
date_applied	Fill in when you apply
referral	Referral contact
recruiter_contact	Recruiter info
followup_date	When to follow up
priority	High / Medium / Low
notes	Source platform, description excerpt
job_category	coding / noncoding

---

Notifications

Option 1: Telegram Bot (standalone, no OpenClaw needed)

1. Message @BotFather → /newbot → copy the token
2. Message @userinfobot → copy your chat ID
3. Add both to config.json under notifications

Test: python3 scripts/notify.py --test

Option 2: OpenClaw channel

• Set openclaw_channel (e.g. telegram), openclaw_account, and telegram_chat_id
• Routed through OpenClaw's messaging system

---

Platform Notes

• LinkedIn: Best coverage for tech/ML roles; occasionally rate-limited
• Indeed: Strong UK/EU/US coverage, often surfaces roles not on LinkedIn
• Both searched simultaneously via python-jobspy

---

Project Structure

jobclaw/
├── SKILL.md                    # OpenClaw skill definition
├── scripts/
│   ├── setup.py                # First-run setup wizard
│   ├── search.py               # Search engine (jobspy + scoring)
│   ├── tracker.py              # CSV read/write/dedup
│   ├── notify.py               # Notifications (OpenClaw / Telegram)
│   └── run_daily.sh            # Daily automation runner
├── references/
│   ├── screening_guide.md      # Scoring rules, interview types
│   ├── keywords.md             # Keyword lists and search strategy
│   └── commands.md             # All commands reference
└── assets/
    └── config.example.json     # Config template

---

Roadmap

• AI-powered JD analysis (Claude) for richer scoring
• Cover letter generation per job
• Interview prep mode (/prep <company>)
• Export to Notion / Airtable
• More platforms (ZipRecruiter, Google Jobs)
• jobs.ac.uk scraper for academic roles

Contributions welcome — open an issue or PR. The CI pipeline auto-publishes updates to ClawHub on every push to main.

---

License

MIT — see LICENSE

---

Built with ❤️ and job search anxiety. Powered by python-jobspy and OpenClaw.