Agent Skills: LinkedIn Inbound Run

Automatic inbound LinkedIn message processing

UncategorizedID: aaaaqwq/claude-code-skills/linkedin-inbound-run

Install this agent skill to your local

pnpm dlx add-skill https://github.com/aAAaqwq/AGI-Super-Team/tree/HEAD/skills/linkedin-inbound-run

Skill Files

Browse the full folder contents for linkedin-inbound-run.

Download Skill

Loading file tree…

skills/linkedin-inbound-run/SKILL.md

Skill Metadata

Name
linkedin-inbound-run
Description
Automatic inbound LinkedIn message processing

LinkedIn Inbound Run

Runs the LinkedIn Inbound Agent to monitor incoming messages.

When to use

  • "check LinkedIn messages"
  • "run LinkedIn inbound agent"
  • "any new messages on LinkedIn?"

How to execute

Step 1: Check prerequisites

# Check that Chrome is running with debugging
lsof -i :9222
# Should show a Chrome process on port 9222

# If Chrome is not running:
$SCRIPTS_PATH/start_chrome_linkedin.sh

Step 2: Run the agent

Dry-run (no CRM writes, no notifications):

cd $AGENTS_PATH/linkedin-inbound
python3 linkedin_inbound_agent.py --dry-run

Full run:

cd $AGENTS_PATH/linkedin-inbound
python3 linkedin_inbound_agent.py

Notification test:

python3 linkedin_inbound_agent.py --notify-test

Step 3: Check results

# View summary
today=$(date +%Y-%m-%d)
cat $AGENTS_PATH/logs/linkedin-inbound-$today.md

# View error log
tail -20 $AGENTS_PATH/logs/linkedin_inbound_errors.log

# View agent log
cat $AGENTS_PATH/logs/linkedin_agent_log.json | jq '.[-5:]'

Step 4: Notify the user

  • How many messages were processed
  • How many new contacts
  • How many POSITIVE/QUESTION replies
  • Whether manual review is needed

Output

  • Telegram notification in Saved Messages (if there are actionable items)
  • Summary log: agents/logs/linkedin-inbound-{date}.md
  • CRM updates: sales/crm/activities.csv, sales/crm/relationships/leads.csv
  • Staged contacts: sales/crm/staging/linkedin_inbound_new_contacts.csv

Errors

CDP connection failed

ERROR: CDP connection failed: ...

Resolution:

  1. Start Chrome with debugging:
    $SCRIPTS_PATH/start_chrome_linkedin.sh
    
  2. Open LinkedIn: https://www.linkedin.com/messaging/
  3. Check login

Classification failed

Agent automatically switches to fallback classification (rule-based). Check if Claude CLI works:

echo "test" | claude -p --model haiku

Telegram failed

Notification text is saved in the log file even if Telegram is not working.

tail -f /tmp/linkedin-inbound-agent.log

Scheduled run

Agent runs automatically via launchd:

# Status
launchctl list | grep linkedin-inbound

# Stop
launchctl unload ~/Library/LaunchAgents/com.yourcompany.linkedin-inbound.plist

# Start
launchctl load ~/Library/LaunchAgents/com.yourcompany.linkedin-inbound.plist

# View logs
tail -f /tmp/linkedin-inbound-agent.log
tail -f /tmp/linkedin-inbound-agent-error.log

Related skills

  • change-review — if a PR with CRM changes is needed
  • add-lead — add staged contacts to CRM
  • telegram-send — manually send notification