QA that closes the loop
Coldtea runs checks where the agent is working and routes failures back with full context — so a red run becomes the next task, not a stale ticket.
Most teams treat a failing check as a notification: a red dot, a ping, a ticket someone picks up later. By the time anyone looks, the context that produced the failure — the diff, the session, the reasoning — has gone cold.
Keep the failure where the work is
Coldtea runs checks where the agent is already working. When something breaks, the failing output stays attached to the session that produced it, so the next step is obvious instead of archaeological.
A red run becomes the next task
Rather than filing a failure into a backlog, Coldtea hands it back to the agent that shipped the change, with the logs and the surrounding context intact. The distance from "broke" to "fixing" stays short.
It is a small shift in where feedback lives, but it changes how quickly work actually closes.