Researching an idea with an agent

Research the existing Feature so the evidence stays with the idea. An agent can continue from the Feature’s goal, body, attachments, and log instead of relying on notes stored on one person’s machine.

Ask the agent to research

Use the research_idea prompt with the Feature key, or describe an idea if no Feature exists yet. The prompt tells the agent to search first, reuse or create a Feature, read its existing context, research the question, attach findings, update the body, and link sources where useful. It finishes with a log comment and a status line.

For Guide new members through their first project, a useful question is whether a new member can create a project and find the next action without help. That is an example research question, not a report of research that has been run.

Keep each research round on the Feature

Attach each finding as Markdown named <topic>-<yyyy-mm-dd>.md, for example first-project-2026-10-27.md. If you revisit the same topic, use the same filename to replace that attachment with the new round. The Feature body and log keep the changing context. The attachment holds the current full finding.

The attachment in the screenshot is demo content that shows the record format. It does not report a research session.

Add source material and related pull requests as links on the Feature when they help. Keep the body concise. Capture the key facts and decisions, what to do next, and open questions. Put the detailed analysis in the attachment.

End every session with a handoff

On the Feature, add a comment that says what you did, what you found, and what comes next. Then write the status line so the next person sees the latest state at a glance. Call the status update even if the line has not changed. It records that the Feature was checked. Do this at the end of every research session.

Another person or agent can open the Feature, read its attachments and log, and continue from there. See capturing ideas and validating an idea.