Cayden Chik develops doover, a recovery tool for AI agents’ shell commands. Earlier research at MIT’s MSEAS group explored numerical sea-ice modeling and reduced-order forecasting.[1][2][3]
Start with Do-over, undo for AI agent shell commands →
GitHub Background
MIT’s MSEAS group records Chik as a visiting high-school researcher from Hwa Chong Institution in Singapore during summer 2023. His report concerned numerical sea-ice models and reduced-order forecasting. The account describes using Dynamic Mode Decomposition to construct a smaller surrogate model; neural networks were a proposed extension, rather than a completed result.[3][4]
Contributions
doover snapshots affected files before supported agent shell commands execute and keeps an action journal. Its undo mechanism covers file changes that Git may not protect, including untracked files and changes outside a repository. The project integrates with Claude Code’s shell-command hooks rather than relying only on checkpoints for editor-based file changes.[2]
Ideas
The repository distinguishes reversible file changes from effects it cannot undo, including external operations. That boundary matters when an agent uses a shell: recovering local files is different from reversing everything a command may have done.[2]
Tools, writing & talks
Writing, research & talks
Start here
A snapshot and journal layer for recovering file changes made by supported AI-agent shell commands.
Start with the recorded example and the documented limitations. Trace which files are snapshotted and what an undo restores before considering it for an agent workflow; it is not a general rollback for external effects.
Developer[2]Original source