Who it is forA developer whose agent made a bad change several steps ago and who wants to return to a known good state without losing later work.
The problemUndoing agent edits by hand is slow and risky, and a plain copy of the project for each step takes a lot of space.
The outcomeThe project is restored to a chosen Commit on a new Branch, and the earlier Commits stay in the Store for comparison.
What you need first
- A DeepSeek Harness (dsh) setup with the dsh-layerfs plugin, as described in the LayerFS × DSH guide
- A project you want agents to work on
- Read the LayerFS 1.0.0 status note: the Store runs in its Disposable profile with no crash or power-loss durability, so keep a copy of anything you cannot rebuild
Steps
- Set up LayerFS on your project using the LayerFS × DSH guide in dsh/README.md.
- Review the Commits on the Branch in the History tab to find the last good state.
- Fork a new Branch from that Commit, since any Commit can be forked at any depth.
- Have the agent continue its work on the new Branch.
- Merge the new Branch into main only after you have checked its result.
Prompt for your agent
Using the LayerFS repo at https://github.com/Ephemeral-AI-Lab/layerfs, find the last good Commit on <branch name> and fork a new Branch from it. Report the Commit ID you chose and confirm the original Branch still exists. Verify that the new Branch starts with the files from that Commit.
Caveat. Rewinding works through Branches and Commits, and nothing is deleted from the Store, but unmounting removes only the Workspace overlay, and the Store has no crash or power-loss durability.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for Ephemeral-AI-Lab/layerfs and "Rewind a project to an earlier Commit and branch from it"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer