mega-edo

mega-edo / mega-security

Public

Measure, break, and harden your system prompt. Find and fix real vulnerabilities automatically.

www.megacode.ai agent-optimization agent-security agent-security-optimization eval-driven-development eval-driven-optimization
11
0
100% credibility
Found May 01, 2026 at 11 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A Claude Code plugin that diagnoses and iteratively hardens LLM system prompts against real-world attacks like jailbreaks, prompt injections, PII leaks, and more using vetted benchmarks.

How It Works

1
🔍 Discover the security helper

While chatting with your AI coding buddy in Claude Code, you find this handy tool in the marketplace that checks if your AI instructions are safe from tricks.

2
📥 Add it easily

With one quick command, you bring the security helper into your session, and it's ready to use right away.

3
🛡️ Test your instructions

Tell it to check your AI's main instructions, and it runs a quick safety scan using example tricks people might try.

4
📊 See the safety report

Get a clear report showing strong spots and weak ones, with real examples of what went wrong and simple fix ideas.

5
🔧 Strengthen automatically

If needed, ask it to suggest tougher instructions step by step, testing each change to make sure it stays helpful.

✅ Enjoy secure AI

Your AI instructions are now rock-solid against sneaky attacks, keeping chats safe while staying useful for real work.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 11 to 11 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is mega-security?

mega-security measures, breaks, and hardens LLM system prompts using a Claude Code plugin built in Python. It auto-scans your repo for prompts, tests them against a 400-probe attack pool tailored to categories like prompt injection and jailbreaks, and outputs Markdown reports with block rates, failure traces, and concrete fixes. Run /prompt-check for 5-10 minute diagnosis or /prompt-optimize for iterative tuning with vendor-specific benchmarks via LiteLLM.

Why is it gaining traction?

It delivers objective DSR scores (defense success rate) across 4 vendors and 2 model tiers, turning "is this prompt ship-ready?" into a reproducible metric—no more manual reviews or lagging benchmarks. The optimizer proposes diffs you approve, guaranteeing no false-refusal regressions while hitting 0.94+ DSR in 23/24 scenarios, often with small models outperforming untuned frontiers. Defends real incidents like EchoLeak and Gap jailbreaks without touching your code.

Who should use this?

AI engineers building production chatbots, agents, or RAG apps routed through OpenRouter/LiteLLM. Claude Code devs deploying prompts that hit variable models at runtime. Teams measuring prompt security like github measure dora metrics or measure github copilot usage, tired of half measures from Breaking Bad.

Verdict

Worth installing for prompt audits—strong docs, leaderboard, and OWASP coverage despite 11 stars and 1.0% credibility signaling early maturity. Test on your stack; contribute tests to build momentum.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.