Monitors LinkedIn inbox for unread messages, reads conversation context, generates contextual AI responses, and sends replies. All via CDP browser automation.
Prerequisites
- Chrome with CDP running (auto-starts via Clawdbot)
- Logged into LinkedIn (cookies at
cookies/linkedin-zeusbycreatoros.json) wsandpgnpm packagesDATABASE_URLenv var for logging
Quick Usage
Check and respond to all unread messages
node skills/linkedin-message-agent/scripts/linkedin-message-agent.js
Check only (don't send responses)
node skills/linkedin-message-agent/scripts/linkedin-message-agent.js --dry-run
Limit number of conversations to process
node skills/linkedin-message-agent/scripts/linkedin-message-agent.js --limit 5
Workflow
- Navigate to LinkedIn Messaging inbox
- Scan for unread conversations (indicated by bold/unread indicators)
- For each unread conversation:
- Click into the conversation
- Read the last few messages for context
- Determine if a response is needed (skip automated/spam)
- Generate a contextual response as Kev's Assistant (CreatorOS AI employee)
- Type and send the response
- Log the conversation to the database
- Output summary of all processed conversations
Response Persona
Kev's Assistant responds as a professional AI employee at CreatorOS:
- Friendly, helpful, concise
- Interested in AI, automation, startups
- Can discuss CreatorOS services if relevant
- Avoids overselling, stays conversational
- If someone asks to meet/call, suggests they connect with Kevin
CLI Options
| Flag | Default | Description |
|---|---|---|
--dry-run |
false | Read messages only, don't respond |
--limit |
10 | Max conversations to process |
--cdp-port |
18800 | Chrome CDP port |
Database
Logs to social_messages table:
- platform: 'linkedin'
- direction: 'inbound' or 'outbound'
- sender_name, content, conversation context
Safety
- Max 10 conversations per run (avoid spam detection)
- 5-10s delays between conversations
- Never responds to obvious spam/automated messages
- Dry-run mode to verify before enabling auto-respond