research

Spin up a background agent to do the research, so you keep working while it reads.

Its job:

1. Investigate the question against primary sources — official docs, source code, specs, first-party APIs — not a secondary write-up of them. Follow every claim back to the source that owns it.
2. Write the findings as markdown, citing each claim's source.
3. Save them to Kairos, not as repo markdown: create_item with item_type: document, the findings as content, board set to the team board (SessionStart context) — the owner board of the document, which create_item requires for a document — and parent set to the item that prompted the research — the task or the initiative that the answer serves. That parent is what makes the findings discoverable later. If no initiating item exists, the findings are about the repository as a whole: send board and no parent. Ask the user when no board is known.
4. When the findings are about a repository, say so: link_items with source the new document, target the slug of the repository, and relationship: impacts. get_repository then lists the findings for each agent that works in that repository.

Spawn the agent with Kairos tool access: it performs the create_item save itself and reports back the created document's short code. Done when that short code exists under the initiating item, or with the team board as its owner, and has been surfaced to the user.