Skip to content
Agent Landscape
Open source · Apache-2.0

Guardrails for
Intelligent Agents.

Minimize token usage, incidents and AI psychosis by richer context, better guardrails and stronger observability for Agents.

Works with

Claude CodeCodexGitHub Copilot CLIGemini CLIVS CodeGitHub Actions

What is Crane

A prompt is not a rule. A contract is.

Agents edit faster than anyone can review. Crane turns what may change into signed contracts and enforces them where the agent acts.

01

Preserve and target

Lines an agent must not touch, and lines it must change. Markers move with the code.

02

Checked before it runs

Every tool call is decided in the agent's hooks. Shell bypasses are caught after.

03

Signed evidence

Hash-chained records show what each agent did, what was denied, and why.

Product

One rulebook, three ways in.

CLI

The rules and the referee

One binary creates signed contracts, tests them, and enforces them in the agent's hooks.

policy payments {
preserve K7M2P9RX;
target 4QZ8M2LD;
}
See the commands →

Extension

Governance in the editor

Select code and press Ctrl Alt F. Zones, flows, and selections as a live graph.

Get it on the Marketplace →

Dashboard

Every teammate's agents, one view

Sign in with GitHub to see decisions, denials, incidents, and tokens. Nothing to host.

Open the dashboard →

Get started

Running in a minute.

Loading the latest release…

Download
Install

      
Then, in your repository
crane repo --https https://github.com/OWNER/REPO
crane init
crane checkpoint baseline
crane agent install --profile claude

Checksums and the extension's .vsix are on the release page.

Reference

Every command, one click away.

    Feedback

    Tell us what to build next.

    Crane in numbers

    Downloads

    —

    across every release

    Visits

    —

    to this page

    Releases

    —

    shipped so far

    Agents

    —

    supported out of the box