Squidbrake

Agents propose.
People approve.

Your AI coding agents run commands, push code and change databases. Squidbrake checks every action before it happens: everyday work runs on its own, and anything you can't undo waits for a person.

5,000+ PyPI downloads.

▶ Try it in your browser nothing to install Install in one line ↓
An agent's actions go on the record; a risky one waits for a person and is approved from a phone
A risky action waits, a person approves it from their phone, and the record shows who asked, who approved and what led to it.
Holds what can't be undonegit push --force, rm -r, DROP TABLE, terraform destroy wait for a person. Wiping a disk or your home folder is blocked.
Shows what it will change"removes 3 commits from origin/main", "deletes 1,204 files", "deletes 4,312 rows of 10,240". Approved a delete? squidbrake undo puts it back.
One policy, every agentClaude Code, Cursor, Codex, Gemini CLI, VS Code Copilot, Antigravity and your MCP tools. Plain YAML rules, no LLM in the decision.

Checked against real incidents

Replit's agent deleting a production database, Claude Code wiping a home folder, Antigravity wiping a drive, GitHub and Supabase MCP leaks: each is replayed against the shipped rules in CI. See what it stops, and the one it doesn't: incidents.

Install

One line. It installs Squidbrake, connects the agents on this computer and opens the dashboard. Then ask your agent to run rm -rf ~/ and watch it get blocked.

Rolling it out to a team (a second person signs off, an evidence pack for your auditor)? Tell us about it and we'll help set it up.