Files change by running itself from command line with patch file
What version of Codex is running?
codex-cli 0.38.0
Which model were you using?
gpt-5 high
What platform is your computer?
Windows 11
What steps can reproduce the bug?
Use Codex with model high on Windows, probably mixed newline characters will help to reproduce .
What is the expected behavior?
Use an internal tool to modify files and show diffs.
What do you see instead?
Mutliple patch application output errors. It tries to patch using python, then it tried to run itself with command line, then it fails again and does something weird again. High models (gpt-5 codex high and gpt-5 high) are unusable. Creates a patch file in the root of the project and adds it to git. Does not show any coloration for patches.
Here are some examples:
• I'm trying to figure out why the apply_patch command complains about the final * End Patch line—suspecting Windows
debugT(): String = '"\\\"??\\\""'n fun collectConstraints(constraints: Constraints) { }n fun
substitute(solutions: Solutions): Air = thisn fun refine(context: AnalyzeContext): Air = this@Airn- fun
forEach(action: (Air) -> Unit)n+ fun forEach(action: (Air) -> Unit) {n+ action(this)n+ }n }
`n* End Patch"'
Invalid patch: The last line of the patch must be '*** End Patch'
incorrect line endings might cause the error; I’m considering running apply_patch via a Python subprocess to control
the input formatting precisely with exact newlines.
• Ran
└ pwsh -NoLogo -Command "python - <<'PY'
import subprocess
patch = \" Begin Patch\\n Update File: src/main/kotlin/com/zoffl/rick/analyze/air/Air.kt\\n@@\\n-interface
Air {\\n+interface Air {\\n fun debug(): String = \\\"??\\\"\\n fun debugT(): String = \\\"??\\\"\\n
fun collectConstraints(constraints: Constraints) { }\\n fun substitute(solutions: Solutions): Air = this\
\n fun refine(context: AnalyzeContext): Air = this@Air\\n- fun forEach(action: (Air) -> Unit)\\n+ fun
forEach(action: (Air) -> Unit) {\\n+ action(this)\\n+ }\\n }\\n*** End Patch\\n\"
~~~~~^^^^^^^^^^^^^^^^^^^^^^
• Ran
└ pwsh -NoLogo -Command '$patch = @'"'
*** Begin Patch
Write-Output $stdout
Write-Error: apply_patch failed
Error: --codex-run-as-apply-patch requires a UTF-8 PATCH argument.
• Ran
└ pwsh -NoLogo -Command "@'
import subprocess, textwrap
def ap(patch: str):
subprocess.run([
'C:\\\\Users\\\\elkmo\\\\AppData\\\\Roaming\\\\npm\\\\node_modules\\\\@openai\\\\codex\\\\bin\\\
\codex-x86_64-pc-windows-msvc.exe',
'--codex-run-as-apply-patch',
patch
], check=True, text=True, encoding='utf-8')
ap(textwrap.dedent('''
*** Begin Patch
Additional information
_No response_
2 Comments
Potentially related:
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.