Permission denied error and model gets confused

[antoyo]

Describe the bug

Hi.
Sometimes I will get a "Permission denied" error when selecting "No, and tell Copilot what to do differently".
This seems to only happen with a few OpenAI models including gpt-5.4 (medium), gpt-5.2-codex, gpt-5.2 (high) and gpt-5.3-codex; Opus models do not seem to have this issue.

Affected version

v0.0.384 to v0.0.423, possibly more.

Steps to reproduce the behavior

A possible reproducer for this bug.
With this file:

fn main() {
  let var;
      let var2;
}

Prompt: Manually align this code.
Then, when seeing the diff, chooses "No, and tell Copilot what to do differently" and say: "Align with 8 spaces".

I got:

Failed to apply patch: Error: Permission denied: Update file /home/user/tests/copilot-bug/src/main.rs

and then the model goes off the rail:

◐ Inspecting permissions

  I'm facing a "permission denied" issue, and I'm considering whether I need to use bash with Python or perhaps a command like `cat > file` to address it. I should check
  the file permissions to see if it's writable. Maybe the `apply_patch` function lacks the necessary permissions, but bash might not have the same issue. I think I need to
  try using a bash command, like `perl -0pi`, but first, I should clarify my intent with the tool.

And it can continue for a while trying to investigate the "permission issue" but will never consider the user feedback.

Expected behavior

Choosing "No" should not cause a "Permission denied" error.
It should always cause The user rejected this tool call. User feedback: Align with 8 spaces (I sometimes get this with this reproducer, so try multiple times if you cannot reproduce).

Additional context

A workaround for this is to use --allow-all-paths and typing your feedback in the prompt after the task is over, but then you don't get to manually approve every edit.

Thanks to fix this issue.

[examon]

This has been fixed in v1.0.11. When rejecting a file edit via "No, and tell Copilot what to do differently", the CLI now correctly passes your feedback to the model instead of throwing a "Permission denied" error. If you're still seeing this on v1.0.11 or later, please feel free to reopen.