---
title: "Comment by Mahdi Jabinpour on \"The agent kept polishing after the task passed, and buried the fix under unrelated refactors\" | Hashnode"
description: "The separate-commit rule is especially useful: optional cleanup stays reviewable and reversible instead of borrowing approval from a correct bug fix. I would ad"
url: "https://hashnode.com/posts/agent-keeps-polishing-after-task-passed/6adcf8a12a992bef4f8d24ad/comment/6add08f3a1ede131d19cc597"
---

[Discussion](https://hashnode.com/posts/agent-keeps-polishing-after-task-passed/6adcf8a12a992bef4f8d24ad)

Comment [The agent kept polishing after the task passed, and buried the fix under unrelated refactors](https://hashnode.com/posts/agent-keeps-polishing-after-task-passed/6adcf8a12a992bef4f8d24ad)

[Mahdi Jabinpour](https://hashnode.com/@mahdi-lumowp)

[Oct 24](https://hashnode.com/posts/agent-keeps-polishing-after-task-passed/6adcf8a12a992bef4f8d24ad/comment/6add08f3a1ede131d19cc597)

The separate-commit rule is especially useful: optional cleanup stays reviewable and reversible instead of borrowing approval from a correct bug fix. I would add a small exception path to the stop rule, though: if the minimal fix exposes a real dependency, report that evidence before broadening scope. Otherwise a file-count boundary can hide a necessary change.
