Describe the bug
I told Copilot (GPT-5.4) to perform a rebase, and it used the apply_patch tool to edit the rebase script. So far, so good. But it wanted to make an incorrect edit. So I denied the tool request and told it what to do instead. However, it saw the tool failing, it claimed that it failed "due to some OS error" and instead tried to write a Python script to make the exact same change.
Affected version
1.0.9
Steps to reproduce the behavior
- use
/model to change to gpt-5.4
- tell it to perform an interactive rebase, starting by setting Git's sequence editor to
sed -i 1ib
- tell it to insert a break somewhere by editing the file whose path is the output of
git rev-parse --git-path rebase-merge/git-rebase-todo
- deny the edit, telling it instead specifically which line number to insert before
- notice how it's trying a different tool to make the same change
Expected behavior
The model should take the feedback as the course-correction it is, not mistake it for an unspecified "OS error".
Additional context
No response
Describe the bug
I told Copilot (GPT-5.4) to perform a rebase, and it used the
apply_patchtool to edit the rebase script. So far, so good. But it wanted to make an incorrect edit. So I denied the tool request and told it what to do instead. However, it saw the tool failing, it claimed that it failed "due to some OS error" and instead tried to write a Python script to make the exact same change.Affected version
1.0.9
Steps to reproduce the behavior
/modelto change to gpt-5.4sed -i 1ibgit rev-parse --git-path rebase-merge/git-rebase-todoExpected behavior
The model should take the feedback as the course-correction it is, not mistake it for an unspecified "OS error".
Additional context
No response