When working with sequential agent sessions:
- Started an initial editing thread/session for bootstrapping (3 editing rounds to reach MVP)
- Moved on to 3+ additional sessions to refine the scripts
- Returned to the first thread which still had non-accepted/pending changes
- Accepted those pending changes to clear diff highlights
- Result: ALL files reverted to the state from step 1, completely losing the MVP work from subsequent sessions
STR:
- In an empty workspace, ask the agent to "create a Hello World one-liner markdown file".
- Don't accept the changes, just go to a new session and say "change Hello World to Hello Harald".
- Accept the changes, go back to the first session and accept the changes there too.
- 🐛 "Hello Harald" is gone
Expected Behavior:
There should be versioning/dirty-tracking to prevent accepting stale changes that would overwrite newer work or mark the old session as applied since more changes got stacked on top of them.
Current Behavior:
No guard or warning exists. Accepting old pending changes silently overwrites all subsequent work, causing data loss.
When working with sequential agent sessions:
STR:
Expected Behavior:
There should be versioning/dirty-tracking to prevent accepting stale changes that would overwrite newer work or mark the old session as applied since more changes got stacked on top of them.
Current Behavior:
No guard or warning exists. Accepting old pending changes silently overwrites all subsequent work, causing data loss.