What this agent does
Turning a research question into something you can actually act on usually means twenty open tabs, half-read articles, and a blank doc at the end. This agent runs the research and hands you a finished briefing — synthesized, cited, and structured to read in five minutes.
stage-1-gather— runs a structured multi-query search across the web (usingweb_search/web_research) to collect the most relevant, credible sources for your question, filtering out SEO filler and low-quality results.stage-2-synthesis— reads every gathered source and writes a structured briefing doc: an executive summary, key findings grouped by theme, points of disagreement between sources, and a citation for every claim.
Every run saves the finished briefing as a dated Markdown file, so you build a running research archive instead of losing the output in a chat scrollback.
Requirements
- Runs on free, built-in research tools (
web_search,web_research) — no paid SERP or research API required. - Works on OpenClaw and on Claude AI (claude.ai with web search enabled, or Claude Code) — the pipeline has no platform-specific dependencies.
- No API key beyond your existing OpenClaw or Claude access.
Setup
- Unzip the package into your workspace, keeping the folder structure intact.
- (Optional) Set a default output length, tone, and citation style so you don't have to restate them every run.
- Run a starter prompt with any research question to see the full pipeline end to end.
- For recurring topics (e.g. a weekly competitor scan), schedule it with a fixed question.
What you get
- Multi-query research across the web that filters out SEO filler and low-quality results before you ever read a source
- A structured briefing with an executive summary, themed findings, points of disagreement, and a citation for every claim
- A dated Markdown archive of every briefing, so past research stays searchable instead of lost in chat history
- Setup that works out of the box on both OpenClaw and Claude AI, no paid research API required
