Social Intel Hub

Find and score engagement opportunities across the web. Surfaces conversations where showing your work is genuinely helpful, not spammy.

How It Works

1. Search — Query keywords from config/keywords.json via web_search tool (Brave)
2. Score — Run results through src/scanner.mjs scoring engine (relevance, freshness, authority, context fit)
3. Suggest — Generate response guidelines via src/responder.mjs matched to your skills inventory
4. Review — Send top opportunities to Telegram for human approval

Running a Scan (Agent Cron)

The scan runs as an OpenClaw cron job with agentTurn. The agent:

Step 1: Search keywords

Load keywords from config/keywords.json. For each keyword, use web_search with freshness: "pw" (past week).

web_search({ query: keyword.term, count: 5, freshness: "pw" })

Rotate through 3-4 keywords per scan (not all 8). Track which were searched last in a state file.

Step 2: Format results as JSON

Structure results as:

[
  {
    "keyword": "AI agent tools",
    "results": [
      {"title": "...", "url": "...", "description": "...", "domain": "...", "published": "..."}
    ]
  }
]

Step 3: Score

Pipe JSON into the scanner:

echo '$JSON' | node ~/Github/axiom-public/agent-tools/skills/social-intel-hub/src/scanner.mjs 4 10

Step 4: Report

Send results to Telegram. Only report opportunities scoring 5+.

If no opportunities found, reply with just a summary line. Don't spam with empty results.

Scoring Breakdown (0-10 scale)

Factor	Range	What it measures
Relevance	0-3	DeFi, agents, onchain, automation keywords
Freshness	0-2	Last 24h = 2, last week = 1, older = 0
Authority	0-2	Domain reputation (GitHub/dev.to = high)
Context Fit	0-3	Match against skills inventory

Final score = raw × keyword weight. Keywords like "LP management bot" have 1.5x weight.

Files

• config/keywords.json — Search terms, weights, exclusions
• config/skills-inventory.json — Your tools/skills for context matching
• src/scorer.mjs — Scoring engine
• src/scanner.mjs — Result processor (stdin JSON → ranked output)
• src/responder.mjs — Response suggestion generator
• scripts/scan.sh — CLI wrapper (--demo for test run)

Engagement Rules

These are baked into the responder but worth repeating:

1. Lead with value, not links. Solve their problem first.
2. Show your work. Link to actual code, not landing pages.
3. Read the room. Don't reply to 3-day-old threads.
4. One reply per thread. Never double-tap.
5. Skip if someone already answered well. Don't pile on.

Adding Keywords

Edit config/keywords.json:

{"term": "new keyword", "category": "defi", "weight": 1.5}

Higher weight = opportunities with this keyword rank higher.

Adding Skills to Inventory

Edit config/skills-inventory.json when you ship new tools. The context matcher uses this to identify threads where your tools are directly relevant.