openai
codex should NOT simplify anything unless asked
Codex keeps implementing stuff "simplified" although explicitly ordered not to / with rules not to in AGENTS.md
It will still often say "I am simplifingy this..", "I am implementing this simplified..."
Codex should ONLY ever simplify stuff if explicitly asked to do so.
Out of curiosity, within your conversation did you at any point use the word 'simplify' or 'let's simplify' something along those lines where you're asking it to implement a simpler solution? If so, these models have a tendency to 'recall' past instructions and will continue to apply those in newer prompts even if you specify it should no longer do this. I accidentally once said something along the lines "Let's discuss this first, no coding..." and then ever after explicitly prompting it to "implement the plan - you're in complete control" etc, it continued to share a plan in future prompts instead of carrying out the work. I had to /new eventually.
Codex doesn't change the context during the conversation. The conversation is only changed when /compact is executed.
The model is very steerable and would pick up on subtle instructions, are you sure there are no mentions of simplification in AGENTS.md?
Thanks for the bug report. We've shared the feedback with the team members who are responsible for training our codex models. We're tracking this feedback and similar reports so we can make the model more robust in future iterations.
If you see other model behavior that you'd like to report, please use the recently-added /feedback command in the CLI. That will provide us with additional details including logs and conversation text.
