Patchworks Health Monitor Agent

The answer is read off that data rather than off a fixed template. The agent locates the failing runs, establishes when they failed, how many times, and whether the failure is still recurring, and reads the logs and scripts behind them. It separates the underlying reason from the error groups that sit downstream of it, so the reply is the one thing that needs fixing rather than every message the flow emitted, and it points out where to investigate next, including practical constraints such as how long payloads are kept.

Where there is a remediation it can describe, it sets out exactly what it would change and why, and asks before anything is applied. An approved change is handed to the Implementation Agent, which edits the flow, for example by inserting a validation step ahead of the step that was failing. The agent is explicit that a guarded fix addresses the symptom and that the upstream cause should still be traced. Flows themselves run on the platform's deterministic integration runtime, so a fix is a configuration change to a flow rather than generated code.

The submission's worked example is a question about recent flow failures. The agent returns one failure in the previous seven days, identifies an invalid JSON payload at the Map step as the root cause, explains that the other error groups are downstream consequences, and says where to investigate next.