Hey, thanks for the report. This is a known limitation. The review file list only tracks changes made through the agent’s built-in edit tools. When the agent decides to use a Python script (or sed/cat) in the terminal for batch operations, those changes bypass the tracking system and won’t show up in the summary.
Related report: Code editing in Ask mode, sed edits escape versioning
As a workaround, you can ask the agent in your prompt to use native file edits instead of scripts, for example add an instruction like “edit files directly, do not use scripts for batch modifications”. This doesn’t guarantee 100% results, but it usually helps.
The team is aware of the issue. Let me know if you have any questions.