Production Error Sweep Loop

Loop Engineering — keep an autonomous responder working the error queue, one root cause at a time, until production is quiet.

🎯 Goal

Continuously pull the top production error, find its root cause (not just the symptom), ship a tested fix, and verify the error stops recurring.

⏱️ Schedule

Suggested interval: 15m (or trigger on new error alerts).

▶️ Run it

/loop 15m "Take the highest-impact unresolved production error. Reproduce it, find the root cause, write a regression test, fix it, and open a PR. Continue until no high-severity errors remain."

🔁 Iteration steps

1. Perceive — read the top error by impact (frequency × severity).
2. Reason — reproduce and trace to the true root cause with the debug agent.
3. Plan — design a minimal fix plus a regression test that fails first.
4. Act — implement the fix; the incident-responder agent documents impact and mitigation.
5. Observe — confirm the regression test passes, then open a PR with /create-pr.

🛑 Stopping condition

The high-severity error queue is empty and every fix is covered by a passing regression test.

🧩 Referenced components

• agent:expert-advisors/debug — root-cause analysis.
• agent:security/incident-responder — impact assessment and mitigation notes.
• command:git-workflow/create-pr — ships each fix for review.

💡 Example

A recurring null-pointer crash is traced to an unvalidated webhook payload; the loop adds a failing regression test, fixes the validation, verifies, and opens fix: validate webhook payload before parse.