codex - 💡(How to fix) Fix Permissions system doesn't support compound commands [1 comments, 2 participants]

TL;DR

The permission system in Codex CLI needs to be updated to support compound commands or automate permission rejections to instruct the model to not use them.

Guidance

• Review the AGENTS.md file to ensure that the prompting to not run compound commands is correctly implemented and consider adding additional guidance on handling compound commands.
• Investigate the possibility of approving compound commands if all individual commands are approved, as a potential simple fix.
• Consider implementing a feature to auto-reject compound commands and instruct the model to run them as separate, individual commands.
• Look into the permission system used by Claude code as a potential reference for updating the Codex CLI permission system.

Example

No code snippet is provided as the issue does not contain sufficient technical details to create a specific example.

Notes

The issue lacks specific technical details about the permission system and the model's behavior, making it difficult to provide a more detailed solution. The guidance provided is based on the information given and may need to be adapted to the specific implementation of the Codex CLI and the gpt-5.4 model.

Recommendation

Apply workaround: Implement a feature to auto-reject compound commands and instruct the model to run them as separate, individual commands, as this seems to be a potential simple fix to make the permission system usable with the model.