Clean My Output

Let the model remember the correction—without making every reader see it.

A ChatGPT Work / Codex lifecycle hook for a specific failure mode: after a user corrects the model, the correction itself keeps leaking into later responses, documents, comments, filenames, and release text.

It remembers each correction for the rest of the session, then checks later replies and writing actions so readers see the result—not the cleanup process. Important safety or limitation notes can still appear once at the end.

Fast evidence

You ask GPT-5.6-sol for a grilled cheese sandwich. It stuffs a whole roast turkey inside. You take the turkey out, then ask for the final menu copy. The correction should shape the result—not become part of it.

The correction stays active for the session. The hook checks final replies, added file text, new filenames, and supported write or publish actions.

Equip it to ChatGPT Work or Codex

Send this prompt:

Install Clean My Output in my current project:
https://github.com/Byte-Naut/clean-my-output

Merge it with existing hooks. Do not overwrite them.

License

This project is released under the MIT License.

Friends: LINUX DO