Your AI agent can start from the wrong version of your code.
It may still write code, pass tests, and look successful—while quietly leaving out work you already finished. Lineage Guard checks first and stops the agent before that happens.
New installations are temporarily paused. We found a situation where version 11 could protect the wrong code project. We are verifying the safety update before reopening setup.
Already installed? Your existing protection checks stay active. Read the update.
$ lineage-guard guard
Nothing is deleted. The agent is simply told to restart from the right version.
Remember the good version
You mark the complete, approved version of your project.
Check before work starts
Every agent is checked against that known-good version.
Stop the wrong version
If approved work is missing, the agent is blocked—not trusted to guess.
The problem in plain English
Imagine two people editing different copies of the same document.
One copy contains yesterday’s approved feature. The other copy does not. An agent opens the second copy, adds something new, and calls it finished. The new work is there—but yesterday’s feature has vanished.
That is what can happen with branches, worktrees, and parallel coding agents. A newer file is not always the complete file.
When do I need this?
Use Lineage Guard when an AI is changing code in a Git project.
Lineage Guard sits inside a software project. It checks the version your coding agent starts from before the agent edits files. It is not something you add to an ordinary chat, document, or website you are only visiting.
- Codex, Copilot, Cursor, or Claude Code is editing files in your repository.
- Several agents, branches, or worktrees are changing the same product.
- Your team needs to ensure every new task includes the work already approved.
Example: You ask Codex to add checkout to your app while another agent has just finished login. Lineage Guard stops Codex if its copy is missing the login work.
- You are having a normal AI conversation without asking it to edit a codebase.
- You are writing text, researching, or viewing a finished website.
- There is no Git repository or coding agent working on project files.
Example: This conversation about Lineage Guard does not need protection by itself. The Lineage Guard software repository does, whenever a coding agent works on its code.
Developers are already describing it
This is not a theoretical problem.
“I wrote all files to the main repo path instead of the worktree path.”Claude Code user on Reddit ↗“
“It lost a whole bunch of prior versions.”Claude user on Reddit ↗“
“Things break badly, including parts of the app that were working before.”Cursor user on Reddit ↗
Lineage Guard does not try to make the AI smarter. It gives Git a simple rule the agent cannot talk its way around.
Easy to install. Hard to copy.
Use it your way. The protection stays with us.
A small connector works inside your project. The actual decision engine runs securely online, so your team gets the result without receiving our core code.
Ask your agent
Paste one prompt into Codex, Copilot, Cursor, or Claude Code.
Use the terminal
Prefer commands? A lightweight connector will take care of setup.
Protect the whole team
The same online check can guard local agents and automated workflows.
Make stale code impossible to ignore.
Simple for your team to use. The decision engine and its rules remain private on our servers.