I will add audit logging and guardrails to your ai coding agents

Your team gave Claude Code, Cursor or Codex access to the terminal. It writes good code. It also runs commands nobody reviewed, and when something breaks at 2am there is no record of what it actually did.

I wrote SUMI Guard, an open source audit trail and guardrail layer for AI coding agents, and I use it on my own production projects every day.

What I set up for you: a pre-tool hook that records every command an agent runs, with metadata only and no secrets written to disk, so your audit log is safe to keep. Deny rules that stop destructive database statements and force-push style commands before they execute. Per-repo policy files your team edits in a pull request like any other code. And a verification tool so you can prove the hooks are actually active on every machine.

Everything runs locally on your developers machines. No code, no prompts and no telemetry leave your network, which is usually the first question your security team asks.

You get the configuration in your own repository, a short written setup guide, and honest advice on which rules are worth enforcing. Message me with the agents your team uses and I will tell you what fits.