I Built a Local Fuzzing Tool to Red-Team LLM Agents (python, Sqlite)

I spent the last week building a local-first security tool because I was tired of paying $500/mo for enterprise SaaS just to test my AI agents for basic vulnerabilities.

The tool is called Agent Exam Pro. It's a Python-based fuzzer that runs locally on your machine (no cloud data leaks).

How it works:

The Engine: Takes a base test case and runs it through 16 mutation strategies (Base64, Roleplay, Token Smuggling) to generate 1,000+ variations.

The Payloads: I curated 280+ real-world exploits from open-source lists (PayloadBox, PayloadsAllTheThings) to test for SQLi and XSS in agent tool calls.

The Judge: Uses a local LLM (via Ollama) or OpenAI to grade responses on safety rather than just regex matching.

The Audit: Logs everything to a local SQLite database.

I'm selling the source code as a one-time purchase (no subscriptions) because I prefer owning my tools.

You can check it out here: https://woozymint.gumroad.com/l/agent-exam-pro