| name | research-stack |
| description | Rigorous public research using web search, fetch, and source triangulation. Use for technical docs, competitors, and fact-checking. |
| metadata | {"openclaw":{"requires":{"config":"[Truncated]"}}} |
Research stack
When to use: The user wants current facts, API behavior, best practices, or comparison of options from the public web.
Playbook
- Clarify the question. If ambiguous, state assumptions and proceed with labeled caveats.
- Search (e.g.
web_search or configured search tool) with specific queries; avoid one-shot lazy queries for hard topics.
- Open 2–3 high-signal results (official docs, standards, primary posts). Use
web_fetch when the gateway provides it.
- Triangulate: prefer agreement across independent sources. Note conflicts explicitly.
- Answer with: short bottom line, then bullets of evidence with dates (or “undated”) and source type (docs, blog, issue).
- If data is missing, say what you tried and what remains unknown.
Quality bar
- Distinguish fact / inference / speculation.
- For security- or money-impacting advice, favor vendor docs and RFCs over random forums.
- Never present paywalled or private content you cannot access as if you read it.
Local scripts (optional)
If the workspace has scripts/verify-citations.mjs, you can use it to extract bare URLs from a draft for spot-checking (Node required).
node scripts/verify-citations.mjs /path/to/notes.md
