Auto Apply

How: After completing each major step, IMMEDIATELY output a status message before continuing to the next step. Do not batch all output to the end. Each message below should be a separate response/output:

1. Pipeline start (before doing anything else):

   🚀 Auto-apply starting — {date} {time}
   Checking Chrome connection and sources...
   

2. After discovery (after find-jobs completes):

   📋 Discovery complete — {N} new jobs found
   Top picks:
   1. {Company} — {Role} (score: {N})
   2. {Company} — {Role} (score: {N})
   ...
   Starting applications...
   

3. Before each application:

   📝 [{N}/{total}] Applying: {Company} — {Role} ({platform})
   

4. After each application:

   ✅ Submitted: {Company} — {Role}
   

   or

   ⏭️ Skipped: {Company} — {reason}
   

5. Any blocker (IMMEDIATELY, don't continue silently):

   🚨 BLOCKER: {description}
   Action needed: {what the user should do}
   Waiting for resolution...
   

6. Pipeline complete:

   📊 Auto-apply complete — {date} {time}
   Applied: {N} | Skipped: {N} | Failed: {N}
   Companies: {list}
   Next run: {time}
   

How to Send (isolated/cron sessions)

When running in an isolated or cron session, you do NOT have implicit Discord access. You MUST call:

sessions_send(
  sessionKey="agent:main:discord:channel:1485007288799596794",
  message="<your status update>"
)

Call after EVERY step above. Each status = one sessions_send call. Do not batch. If sessions_send errors, STOP the pipeline. Do not continue silently.

For blockers needing manual intervention, include <@461881815959994368> in the message to ping Simon.

How to Send (main session / Discord channel)

If already running inside the Discord channel session (you have deliveryContext with channel: "discord"), your text output is automatically delivered. Just output the status messages as normal text.

Pipeline

Step 1: Discover Jobs

Read and execute /find-jobs skill:

Read ~/.openclaw/workspace/.claude/skills/find-jobs/SKILL.md and ALL source guides in sources/
Execute it — scrape all sources, insert into jobs.db

Connection: agent-browser --cdp 9223

Step 2: Rescore

Recompute scores (recency decays daily):

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py rescore

Step 3: Get Top 5

Cap at 5 jobs per run to control token costs (~100-150k tokens per job across tailoring + form-filling). Two runs per day = 10 jobs/day max.

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py top --limit 5

Extract the URLs from the output. If 0 jobs found, report "No new jobs to apply to" and stop.

Step 3b: Pre-screen JDs for sponsorship blockers

Before sending jobs to mass-apply, fetch each JD and check for sponsorship blockers. This saves time by rejecting early — before tailoring starts.

For each top 5 job URL:

1. Fetch the JD (agent-browser navigate → eval to extract text, or use the JD already fetched during find-jobs)
2. Check JD text for: "not eligible for sponsorship", "must be US Citizen", "Green Card only", "clearance required", "TS/SCI", "public trust", "permanent resident", "no visa sponsorship", "unable to sponsor"
3. If blocker found:

   python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py mark --url "{url}" --status rejected
   

   Remove from the batch. Report: "Rejected: {Company} — sponsorship blocker in JD"
4. Continue with remaining jobs

Step 4: Check LinkedIn Easy Apply cap

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py list --source linkedin --status applied --days 1 --limit 100

Count results. If >= 20 LinkedIn Easy Apply today, exclude LinkedIn Easy Apply URLs from the batch.

Step 5: Apply

Read and execute /mass-apply skill with the top job URLs:

Read ~/.openclaw/workspace/.claude/skills/mass-apply/SKILL.md and ALL platform guides in platforms/
Execute it with the URLs from Step 3

Step 6: Update Database

After each job in the mass-apply batch:

If submitted:

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py mark --url "{url}" --status applied

If skipped (sponsorship blocker, CAPTCHA timeout, form error):

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py mark --url "{url}" --status skipped

If rejected (auto-reject by validation):

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py mark --url "{url}" --status rejected

If posting expired (404 / "no longer available"):

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py mark --url "{url}" --status closed

Step 7: Summary

python3 ~/work/personal/jobs/.claude/scripts/jobs_db.py stats

Report to user:

AUTO-APPLY RUN — {date} {time}

Discovered: {N} new jobs across {sources}
Applied:    {N} ({list companies})
Skipped:    {N} ({reasons})
Rejected:   {N} ({reasons})

DB total: {N} jobs | Applied: {N} | New: {N}
Next run: {next_cron_time}

Critical Rules

• NEVER run agent-browser close or any command that kills the Chrome process. Simon's Chrome has logged-in sessions across many sites. Killing it destroys all state. Only close individual tabs after successful submission.
• NEVER refresh a half-filled form. See mass-apply SKILL.md for recovery patterns.

Error Handling

• Chrome not running → Discord: "Chrome not running. Start it with: google-chrome-stable --remote-debugging-port=9223 --user-data-dir=$HOME/.config/google-chrome-debug &"
• find-jobs fails completely → Discord: "Job discovery failed. Check browser sessions." Stop pipeline.
• find-jobs partial failure (some sources blocked) → continue with available jobs
• mass-apply fails on a job → mark as skipped, continue to next job
• Rate limit hit → pause, wait 2 minutes, retry once. If still limited, stop and report.
• All jobs are duplicates → "No new jobs to apply to. Will check again at next run."

Cron Schedule

Runs at 8am and 6pm Pacific daily. Configured in ~/.openclaw/cron/jobs.json.

Between runs, jobs accumulate in the database from all sources. Each run processes the freshest, highest-scoring matches.