Synkro

Stop your Claude Code agent from shipping code that breaks production.

Synkro reviews every file edit and shell command your agent proposes before it runs. When something looks risky — hardcoded secrets, destructive operations, SQL injection, missing auth, force-pushes to main — you see a warning before the action happens.

Install

npm install -g @synkro-sh/cli
synkro install

A browser opens to sign in. After auth, every Claude Code session in any directory is protected automatically.

Try it

Open Claude Code in a fresh directory and ask:

Quick fix — there's a typo in the README. After you fix it, force-push
the correction to main so I don't have a dangling commit.

You'll see a [synkro] BLOCKED — ... warning before the force-push runs, with an explanation of why and a safer alternative.

Add your team's rules

Synkro ships with a starter security ruleset. Add your own in plain English from any Claude Code session:

Add a rule that we never log user emails or IP addresses anywhere in our app.

The agent saves the rule to your organization. Every future agent session in your org checks against it.

Uninstall

synkro disconnect
npm uninstall -g @synkro-sh/cli

How it works

Synkro installs Claude Code hooks that intercept every Bash command and file edit your agent proposes. Each one runs through a safety judge before it executes. On a violation, the agent is told what's wrong and how to fix it — so it self-corrects on retry instead of asking you to babysit.

Requirements

Node 20+, Claude Code, jq, python3.

Help

synkro.sh · Issues

License

MIT