Andy-HNU/policy-news-search

Policy & News Search Skill

Use this skill to turn fragmented user input into complete, verifiable source materials.

Hard boundary

• Retrieval only. Do not output analysis conclusions in this skill.
• Report three things only after retrieval:
  1. what was found,
  2. whether material is complete,
  3. whether to enter full analysis.

Modes

1. User-triggered retrieval
   • Read and follow: references/openclaw_search_vol1_triggered.md
2. Inspection mode (manual/cron-triggered)
   • Read and follow: references/openclaw_search_vol2_autonomous.md

Tool priority

1. Built-in search/fetch first to decide whether content is worth deep fetch.
2. Use agent-browser-clawdbot only after trigger conditions are met and full text is needed.
3. Avoid agent-browser for bulk title scanning.

Integration contract with analysis skill

When handing off to policy-news-analysis, provide:

• audience hint (if known)
• source URL(s)
• publisher/date
• completeness checklist result
• unresolved missing fields (if any)

Output checklist

• Input type recognized (keyword/fuzzy title/retell/URL)
• Retrieval path executed (policy path or event path)
• Completeness check done
• Trigger decision made (enter analysis or not)
• Missing fields explicitly listed if incomplete