Real Plan And Rigorous Mode
What variant of Codex are you using?
codex cli
What feature would you like to see?
Please its a pain. I love codex, chatgpt mentality but he is not rigorous ..
And first codex software doesn't implement plan system like claude code or cursor.
I mean by that. When i do a plan with codex. I Ask codex to ask me some questions, multiple question to polish the direction but each time he construct the plan, he start from zero... Its seem like vibe coding tool and not for software engineer
Each time he construct the plan, he forget some elements of the previous plan ... it make me frustrated
And please fix that : Codex is some time lazy. I make a complete plan handwritten with collaboration of codex on a target file.
He have the directive to implement the plan from the file.
Guess what ?
On 20 tasks, he implement 3 and ask me "if you want i can do that ? " Yes I Want ! but its boring ...
Or sometime you are a bit tired. He says he finished and you go on the project and "same" only 1/3 of details is done.
That's vibe coder stuff.
How can i trust vibe coder agent.
If you change that, i'm sure many people from claude code or cursor will come on codex.
Codex is extremely smart but i think more with time that, vibe coders is killing him. It make me very frustrated that the software is killing the rigorous behavior
AGENTS.md change nothing.
Everything let me think. Your codex software won't let me focus on other tasks.
I repeat the model is EXTREMELY GOOD. You just have vibe coder stuff in it. Kick that.
I want a true collaborator. Sometime i prefer to go back on chatgpt. chatgpt stay remains friendly but i don't know how to say it, he is more like he should "an ambitious and friendly and serious gpt" but codex is more orientating him to a lazy agent
I just want more serious. Plan and tasks system that pilot codex. Codex should not stop the processus
Additional information
_No response_
5 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Theses issues are closes and do not explain the real and deep problem
If you'd like Codex to do deeper analysis during planning, you can steer it in that direction by specifying what you want in the prompt or your AGENTS.md instructions. You can also switch to a higher "reasoning effort" like "high" or "xhigh".
The "laziness" issue that you're referring to is a model behavior, not something that the Codex model has control over. Once again, you can help combat this with prompts.
No you are not right. This isn't an issue of the model. Its an issue of the software.
Its not "i would like deeper analysis during planning" Its i want the model to respect the tasks list.
But its ok. Good Luck
Obviously, OpenAI restricted it on the backend.Its not a bug.