job-claw（工作爪）

JobClaw

AI-powered job search automation. Searches multiple platforms, scores results against the user's profile, and maintains a local CSV tracker. No backend server or web dashboard required.

Quick Reference

See references/commands.md for all CLI commands and chat commands. See references/screening_guide.md for scoring logic and interview type classification. See references/keywords.md for keyword lists and search strategy.

Conversational Setup (First Run)

When a user first activates JobClaw, do not run setup.py directly. Instead, guide them through a natural conversation to collect all config values, then write config.json yourself at the end.

Check first:

test -f ~/Documents/JobClaw/config.json && echo "exists" || echo "not found"

If config doesn't exist, start the conversational flow below. Ask one topic at a time — never dump all questions at once.

Conversation Flow

Step 1 — Greeting

Hey! Let's get JobClaw set up. I'll ask you a few quick questions and have you searching for jobs in minutes.

First — what's your name?

Step 2 — Background (after they give name)

Nice to meet you, [name]! 

Tell me a bit about your background — what kind of work do you do or want to do? 
(e.g. "software engineer with Python/ML experience", "data scientist at a fintech", "researcher in NLP")

This helps me score jobs that actually fit you.

Step 3 — Skills (after background)

Got it. What are your key technical skills or tools? List whatever feels most relevant — I'll use these to boost scoring for matching jobs.

(e.g. pytorch, python, sql, docker, react, java — whatever fits you)

Step 4 — Target roles (after skills)

What kinds of roles are you looking for? You can be specific or broad.

(e.g. "ML engineer or data scientist", "backend engineer", "product manager", "research scientist")

Step 5 — Locations (after roles)

Where are you looking? List cities or regions — I'll search each one separately.

(e.g. "London, UK", "Remote", "Berlin, DE and Amsterdam, NL", "New York, NY")

Step 6 — Search preferences (after locations)

A couple of quick settings:

- How recently posted should jobs be? (default: 48 hours)
- Minimum match score to save a job? (default: 70 out of 100 — higher = fewer but better results)

Hit enter to keep defaults, or tell me what you'd prefer.

Step 7 — Daily schedule (after preferences)

Do you want a daily automated search? I can run it every weekday morning and notify you of new matches.

If yes — what time? (e.g. "7:30am", default is 07:30)
And your timezone? (e.g. Europe/London, America/New_York)

Step 8 — Notifications (after schedule)

Last one — how do you want to be notified when new jobs are found?

1. Telegram Bot (you'll need a bot token from @BotFather)
2. Through OpenClaw (if you're using another channel like Signal or WhatsApp)
3. No notifications — I'll just save to CSV

Which works for you?

For option 1, ask for: bot token + chat ID (tell them to message @userinfobot to get their chat ID). For option 2, ask for: channel name (e.g. telegram) and their chat/user ID. For option 3, set notifications.enabled: false.

Step 9 — Write config and confirm

Once all answers collected, write ~/Documents/JobClaw/config.json using the template from assets/config.example.json, populated with the user's answers. Then confirm:

All set! Here's your config summary:

👤 Name: [name]
🔍 Skills: [skills]
📍 Locations: [locations]
⏰ Daily search: [time], [timezone]
🔔 Notifications: [method]

Want me to run your first search now? (takes ~1-2 minutes)

If they say yes, run: python3 <skill_dir>/scripts/search.py --mode all

Daily Automated Search Workflow

Triggered by cron or the user saying "run job search":

# Reads: ~/Documents/JobClaw/config.json
# Writes: ~/Documents/JobClaw/data/jobs.csv
# Logs:   ~/Documents/JobClaw/logs/daily.log

./run_daily.sh              # both coding + noncoding
./run_daily.sh --mode coding
./run_daily.sh --mode noncoding

Or directly:

python3 scripts/search.py --mode all
python3 scripts/search.py --mode coding --dry-run

The search pipeline:

1. Reads config.json for keywords, locations, min_score, platforms
2. Searches LinkedIn + Indeed via python-jobspy
3. Scores each job (keyword-based, 0-100) using user.skill_keywords as boosters
4. Filters by min_score (default 70), deduplicates by company+role and URL
5. Appends qualified jobs to data/jobs.csv
6. Sends notification via OpenClaw/Telegram (if configured)

/newjob Command

When user says /newjob <url>:

1. Fetch the JD: browser(action="navigate", url=<url>) then browser(action="snapshot")
2. Analyse against references/screening_guide.md:

- Company, Role, Location, Work Mode, Salary - Match Score 0-100 with reasoning - Interview Type estimate - ML Direction, Seniority

1. Add to CSV:

   import sys; sys.path.insert(0, "<skill_dir>/scripts")
   from tracker import JobTracker
   t = JobTracker()
   t.add_jobs([{...job dict...}])

1. Reply with score, match reasons, interview type, apply angle

Archive Management

When user says "archive expired jobs", "clean up old jobs", "auto-archive":

# Dry-run (see what would be archived, no changes)
python3 scripts/archiver.py

# Apply archiving
python3 scripts/archiver.py --commit

# View archive stats
python3 scripts/archiver.py --stats

# Restore a job (by company name, role, or URL fragment)
python3 scripts/archiver.py --restore "DeepMind"

Archive rules (same as the web dashboard):

• expired_30d — status New/Interested with no update for 30+ days
• auto_rejected — status Rejected/Passed for 60+ days
• url_dead — non-LinkedIn URL returns 404/410 or contains "job has expired" etc.

Archived jobs move from data/jobs.csv → data/jobs_archive.csv. Auto-archive runs automatically at the end of every daily search (run_daily.sh).

Showing Stats

When user asks "show job stats", "how many jobs", "top jobs":

import sys; sys.path.insert(0, "<skill_dir>/scripts")
from tracker import JobTracker
t = JobTracker()
t.print_summary()
top = t.top_jobs(10)

Setting Up Daily Automation (OpenClaw Cron)

1. Confirm daily time and timezone from config.json
2. Register an OpenClaw cron job:

   Schedule: daily at <config.schedule.daily_time>, weekdays only
   Command: JOBCLAW_DIR=~/Documents/JobClaw <skill_dir>/scripts/run_daily.sh

Updating Config Mid-Conversation

When user says things like "also search in Berlin", "add React to my skills", "change time to 8am":

1. Load ~/Documents/JobClaw/config.json
2. Edit the relevant field
3. Save back
4. Confirm the change: "Done — I've added Berlin, DE to your search locations."

Environment

• Config: ~/Documents/JobClaw/config.json (override with JOBCLAW_DIR env var)
• CSV data: ~/Documents/JobClaw/data/jobs.csv
• Logs: ~/Documents/JobClaw/logs/daily.log
• Dependency: python-jobspy — install with pip install python-jobspy