Codex

What version of Codex is running?

codex-cli 0.56.0-alpha.5

What subscription do you have?

PRO

Which model were you using?

GPT5/ GPT-5 CODEX

What platform is your computer?

Macbook M2 Pro; macOS Tahoe v26.1

What issue are you seeing?

On VSCode, the codex integration (with paid OPEN AI ChatGPT version)

1. Is unable to resolve paths if it's being told to fetch files from a certain location.
2. Starts making random edits to the source file when it should be making edits to the destination file.
3. Despite the destination and CWD/PWD files being open in the context window, it ends up making edits with a mind of its own to the source files which it was asked to only reference to.

This happens when the 100% quota for a chat is just about to be over and Codex auto resets the request % counter in the same chat.

1. It creates complicated python scripts when simple bash one-liners can do the trick.
2. In LATEX .tex files, it can not escape \t for example, \textcolor is being written as extcolor
3. Hallucinates and starts making edits that were requested 20-25 requests before. Says it's sandboxed to a certain directory despite the right directory being open in the context window. Will consistently lie about it.

What steps can reproduce the bug?

Please refer to the issues I am seeing; it's not a matter of my code, but what Codex is hallucinating. Codex is burning through requests percentage unnecessarily when , say, Claude will do it in 3 quarters of the time.

What is the expected behavior?

None of the reported issues above should exist in the first place. Quite odd; these should be the basic checks before a version is released; seems like untested versions are being released with confidence.

Additional information

_No response_