Agent Activity and Safe Revert
Review attributed external changes and revert only operations that are still safe to undo.
Agent Activity is the trust layer for externally attributed workspace changes.
What appears
When attributed agent sessions exist, the workspace exposes an Agent Activity history containing:
- actor/client name
- last-change time
- operation and batch counts
- created/updated/moved/deleted summaries
- batch labels and concise operation details
- revert result
The control stays absent when there is no agent activity.
Safe Revert
Revert is guarded. Wall walks the agent session backwards and verifies that the current entity still matches what the agent left behind.
Examples:
- agent creates an untouched Note → safe to remove
- agent updates an untouched Note → safe to restore
- agent updates a Note, then you edit it → agent revert skips that conflicting operation
Skipped operations are intentionally kept rather than overwriting later human work.
Normal Undo still exists
The generated revert is itself a normal user operation batch, so it is persisted/synced and can participate in normal Undo.
Across sync providers
Supported agent attribution travels with Noticky Cloud and iCloud entity mutations so another device can reconstruct the same actor/session context.
Public Canvas shares do not publish the local Agent Activity journal.