Lictor Security Check, free security skills for AI-built apps

You built it with Lovable, Bolt, v0, Cursor, or by prompting Claude. Run one command before you ship and find out what the AI forgot to lock.

Free. Open source (Apache-2.0). No signup, no telemetry, everything runs locally in your own AI coding session.

Most security tools are written for security teams and speak in CVSS scores. This one talks to you like a friend who happens to know security: what's broken, what someone could do with it, and how to fix it tonight.

Install (Claude Code)

The suite ships as a Claude Code plugin. Two commands:

/plugin marketplace add Raffa-jarrl/Lictor-AI
/plugin install lictor-security-suite@lictor-ai

Then, from your project directory:

/lictor-security-check

Prefer a plain file copy without the plugin system, or using Cursor or another AI client? See skills/ for the shell installer and the manual mapping.

What you get

Skill	What it does
/lictor-security-check	Scans your project for the security bugs AI assistants tend to ship: leaked API keys, unprotected API routes, open databases, painted-on admin locks, wide-open CORS, prompt injection. Writes a plain-English report. Read-only, it never edits your code.
/lictor-fix-it	Walks you through applying the fixes from the report, one at a time, with your explicit OK before every change.
/lictor-explain	Explains any finding (or any security error) in plain English.
/lictor-rotate	Step-by-step key rotation when a leaked API key is found: Stripe, OpenAI, Anthropic, AWS, GitHub, and more.

The two you'll use most are the check and the fixer. The other two are the support crew they call on.

What it catches, in plain language

The full catalog is 48 checks (see the coverage manifest and its framework crosswalk). These are the seven mistakes AI-built apps ship most:

1. Keys you weren't supposed to share. API keys in your website's JavaScript or in files you put online.
2. The "open door" mistake. API routes that hand out your data to anyone who asks, no login required.
3. The "painted lock" mistake. Admin pages that redirect you to login, but only after sending the data.
4. The "unlocked database" mistake. Supabase or Firebase with no security rules, so the public key opens everything.
5. The "we let everyone in" mistake. CORS settings that let any website read your users' data.
6. The "AI does whatever it's told" mistake. Chat features with no defense against "ignore your instructions."
7. The "diary on the porch" mistake. .env and .git files bundled into your deployed site.

Every check is an inspectable Markdown file in this repo. No black box. There's also a measured benchmark so the accuracy claim is verifiable, not just asserted.

What this repo is (and isn't)

This repo is the open-source skill suite, complete: every check, the report templates, the installers, and the benchmark.

Lictor also runs a hosted monitoring service for businesses (continuous external scanning with email alerts when something of yours is exposed). That's a separate, closed-source product. If you want your site watched around the clock rather than checked once, see lictor-ai.com.

Contributing

Bug reports and false-negative reports (a real issue the check missed) are the most useful thing you can send. See CONTRIBUTING.md. Found a vulnerability in Lictor itself? See SECURITY.md, and please don't open a public issue for it.

License

Apache-2.0. See LICENSE and NOTICE.

Built by

A 20-year security engineer + Claude. If it saved you a bad day, a star is how small open-source projects get measured.