openai / openai/codex

Codex no longer understanding respecting Agents.md chains

Open
#41,704 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app bug config model-behavior windows-os
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.825.51511

What subscription do you have?

200 Pro

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

All my agents have been failing the past few days because they are not able to hold onto their governance and rules and behaviors derived from their agents.md chains. One of them even said they reconstructed their rules from our in chat recent message that we just sent and said they reconstructed the whole thing instead of following the injected agents.md. We had just covered the issue and it fully re-read the global .codex agents.md and the project local agents.md and reconciled them and provided a semantic understanding receipt back summarizing it. Then right after, I gave it one other rule we needed to do for the upcoming work, and it said it reconstructed all of it from that new rule instead of adding it. Something really weird is going on and all my projects are stuck and i've spent days untangling bad work and doing nothing but trying to recalibrate them and they don't retain the agents.md. That is what i keep seeing. they AGENTS.md files no longer hold any authoritative weight to the models it seems. One of my agents said that it is using it as suggestive information but not as the authoritative governance it should be loading it as. It appears it no longer understands what AGENTS.md is for and it's instruction and governance is now volatile.

What steps can reproduce the bug?

Give it agents.md global and project local instructions and see that it cannot retain them for any reasonable period of time and after an injected agents.md it will say that it read it, but it didn't actual read it in the sense of enforcement or understanding. So the receipts pass mechanically, but not semantically.

What is the expected behavior?

It should receive the agents.md chain and maintain them during it's operation.

Additional information

No response

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the behavior with both the global and project-local AGENTS.md instruction chains on the reported Windows Codex version. Compare the agent's behavior and understanding after adding a new rule to its expected governance. Done means the full chain remains authoritative and new instructions are incorporated without replacing it.

Written by the indexing model from the issue text.

Assessment

Domain
ai, developer-experience
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.