Ledgerful – I built a local tool to catch when AI invents stuff in my code

I kept having problems with Agents not considering downstream effects of code changes. So I built Ledgerful, a local CLI/MCP to help. Ledgerful maps the code, estimate blast radius, run structured verification, and keep a local, signed ledger of what was intended and checked.

Ledgerful signs every change you make in a git repo, lets you verify the chain offline, and produces a SOC 2-style evidence export. It runs on your machine. Nothing uploads by default.

Here’s where you can find it: https://github.com/Ryan-AI-Studios/Ledgerful What I am hoping to learn: 1. Is this a problem you have? If yes, what's the most painful part of catching AI-invented code in your workflow? 2. What would the killer feature be that would make this part of your daily workflow? 3. Anything that concerns you about the approach? Signing per commit, local-first, the Action's threat model. All fair game.

Thank you in advance for any input.