Fix Ci
Compose the agent’s fix-CI prompt for the given PR — does NOT send it. The frontend takes the prompt and calls the normal sendMessage flow so the chat UI reacts as if the user typed it.
The composition is server-side because we re-fetch the current failing-check names from GitHub right before building the prompt; that keeps the model’s context fresh and avoids the frontend duplicating the github_service round-trip + failure aggregation logic. The user message that lands in chat IS the prompt below — so the user can read exactly what was asked, and any follow-up turns can refer to it like any other message.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Response
Successful Response
Server-composed fix-CI prompt for the frontend to send as a system-event chat message via the existing useTaskStream.sendMessage flow. Returning the prompt (rather than streaming the message server-side) lets the chat UI render the optimistic chip, stream the assistant chunks live, and persist the exchange via the same code path as a hand-typed message — no separate "did the chat update?" plumbing for synthetic messages.
display is the short chip text the chat should SHOW (persisted as
the message content via TaskMessageCreate.display_content); prompt
is the full instruction only the agent receives. kind is always
"system_event" — echoed so the frontend forwards it verbatim.