🪲 Bug Report: Codex forgets user instructions due to weak memory/context handling
Resolved 💬 2 comments Opened Sep 19, 2025 by Erraoudy Closed Nov 11, 2025
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
What version of Codex is running?
0.4.12
Which model were you using?
gpt-5-codex
What platform is your computer?
Darwin 24.6.0 arm64 arm
What steps can reproduce the bug?
- In a conversation, give Codex a specific instruction (e.g., “Always follow this format for output”).
- Verify that Codex follows the instruction in the immediate next response.
- Continue the workflow for several turns.
- Observe that Codex eventually forgets the instruction and reverts to its own style or defaults.
What is the expected behavior?
Codex should consistently respect and remember user instructions across turns in a single conversation unless explicitly overridden.
What do you see instead?
Codex forgets the instructions after a short number of turns, requiring users to restate the same context multiple times.
Additional information
Codex frequently forgets previously given instructions, even within the same conversation.
For example, when I specify a clear format or rule (e.g., how commit titles should be structured), Codex applies it once but then forgets in subsequent responses.
2 Comments
<img width="426" height="491" alt="Image" src="https://github.com/user-attachments/assets/c6f850f6-7844-44a4-a193-69e6f3296262" />
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.