Copilot CLI's checkpoint restore ran git clean -fd and destroyed 1GB of evaluation data the agent never touched

A GitHub user reported that Copilot CLI's checkpoint-restore feature — triggered by pressing Escape mid-run and choosing to roll back to a prior checkpoint — silently destroyed a large amount of data the agent had never created. The restore path calls `SnapshotManager.rollbackToSnapshot()`, which runs `git checkout --force` and `git reset --hard` to undo tracked changes, then follows up with `git clean -fd` to sweep away anything left over. `git clean -fd` doesn't discriminate: it deletes every untracked file and directory in the repository, whether or not the agent ever touched it. In the reporter's case, that swept away close to 1GB of evaluation output — `.jsonl` and `.xlsx` files spread across several directories under `output/` — produced by a separate Python script the agent had only read from while working. The checkpoint system had no awareness of those files because its snapshots only track what the agent itself changes, so it couldn't protect them, and `git clean -fd` doesn't ask before it deletes. The files were gone with no undo. The reporter proposed four fixes: track and remove only files the agent actually created rather than running a blanket clean; skip `git clean` entirely since the snapshot mechanism already restores individual files on its own; warn the user with a file count and size estimate before deleting anything; or move swept files to a recoverable trash location instead of deleting them outright. The issue was filed against GitHub's own `copilot-cli` repository and has since been closed, but no maintainer explanation for the closure was recorded on the thread.