Hi OpenAI team!
I’d like to thank you all for your amazing work and tools first!
And want to propose improvements for Codex Plan Mode:
- Save somewhere the plan the codex proposed for the first time as a file
- If the user requests changes to the plan, make changes to that file
Current flow: codex stores the proposed plan at the context window, and if I request any changes (even the small ones) and ask it not to change anything else, it still rewrites the plan due to LLM non-determinism.
It’s very hard to go over the plan review again and again. I want to see only the small changes Codex made to our plan, not a rewritten plan from scratch.
Hi, welcome to the forum.
Here are some little tips that I use:
- Once plan mode is ready to implement, hit the stop button/say no.
- At this point you can also change the model/reasoning levels. This way you can make the plan with xhigh reasoning and implement with high/medium reasoning and save usage.
- Prompt this as a follow-up:
Save this plan in myplan.md, break it down into tasks with checkboxes for each task and tick them as you implement them.
- This will also prevent compaction from damaging your plan execution if it is too complex.
- For incremental changes, you can refer your myplan.md and ask codex to update it with new changes, and use git to keep track of what is modified.
For some of my repo’s I add a roadmap table as part of the README.md that way when I feel like it, I can dip back in to a codex session and pick off a roadmap item. This may be too high level for you, but essentially it means that todos are a persistent file!
Usually I try to cover off a whole roadmap item in one session, so persistent “interim” goals aren’t necessary.
The other benefit of this is that blue sky thinking is not lost.
Thank you for taking the time to share this feature request. We appreciate and value your feedback. While we can’t promise implementation or provide a timeline, we’re grateful you shared it with us. We’re closing this topic while keeping it visible for reference.