Looselips Guard — Data skill for Claude Code

Loose lips sink shops.

You want to run agents unattended. Sure, you can sandbox them and run a local model — but to do anything useful you have to let them access the outside world: the internet, git, your email. And nothing there stops an agent firing your personal data when it shouldn't, because the model doesn't know any better.

Existing agent tools focus on stopping your secrets reaching the model. **looselips-guard stops the agent publishing your data to the world.** It blocks a command *before it runs* when what's leaving the machine has something in it that shouldn't go: a banned string from your denylist, a credential, or a suspiciously big file that's probably a database dump. All three run in the one hook, in a single pass over the payload, before the command executes — the credential check is [gitleaks](https://github.com/gitleaks/gitleaks)' 221 secret-detection rules ported to run in-process, not a separate scanner you install or invoke.

It covers the two routes a git-history scanner misses completely:

• issue and PR bodies — they never become git objects, so a scanner never sees them
• git add -A quietly sweeping a live database onto a public branch

---

Getting started

**1. Get `looselips-guard` on your machine.** One of:

npm install -g looselips-guard