Social Inbox

Fetches X/Twitter mentions and brand searches, deduplicates against already-replied tweets, scores by engagement potential, and outputs a prioritized inbox.

Usage

# Full scan: mentions + brand search
node ~/Github/axiom-public/agent-skills/skills/social-inbox/social-inbox.mjs

# Mentions only
node ~/Github/axiom-public/agent-skills/skills/social-inbox/social-inbox.mjs --mentions-only

# Output as JSON (for piping)
node ~/Github/axiom-public/agent-skills/skills/social-inbox/social-inbox.mjs --json

# Custom search terms
node ~/Github/axiom-public/agent-skills/skills/social-inbox/social-inbox.mjs --search "AppFactory,axiom ventures"

Output

Saves to ~/clawd/data/social-inbox.json with structure:

{
  "lastScan": "2026-02-08T...",
  "items": [
    {
      "id": "tweet_id",
      "author": "@handle",
      "text": "...",
      "url": "https://x.com/...",
      "score": 8.5,
      "scoreBreakdown": { "relevance": 3, "question": 2, "directMention": 2, "recency": 1.5 },
      "status": "pending",
      "source": "mentions"
    }
  ]
}

Scoring (0-10)

Signal	Points
Direct @mention	+2
Question mark (asking us something)	+2
Keywords: fund, invest, agent, build, LP, token	+1 each (max 3)
Recency (<1h: +2, <6h: +1.5, <24h: +1)	up to +2
Not a bot/spam pattern	+1 (default, -1 if detected)

Items scoring <3 are filtered out. Already-replied tweets are excluded.

Cron Integration

Add to OpenClaw cron for periodic scanning:

{
  "kind": "agentTurn",
  "message": "Run social inbox scan: node ~/Github/axiom-public/agent-skills/skills/social-inbox/social-inbox.mjs\nReview the top 3 items and draft replies if valuable."
}

Dependencies

twitter-api.py (already installed)
twitter-replied.json dedup log (already exists)
Node.js 20+