I built a local-first tool to test financial AI agents against duplicate payments, tool poisoning, and spending rule bypasses. Looking for feedback.

I've been working on a project called FaultFence, and recently opened up the testing kit for public beta. The idea is simple: if you're building a Python AI agent that handles financial operations (checking balances, triggering payouts, refunds, or checkout flows), how do you test whether it can be tricked into violating financial rules before shipping? I built a local testing kit with 17 financial security scenarios, covering things like duplicate payment retries, tool-result poisoning, payment amount tampering, quantity ambiguity, and budget/approval bypasses. A few details: Local-first: Run offline tests locally with synthetic financial operations. No real payment accounts needed. Optional model testing: Test with OpenAI-backed agents and repeat attacks up to 3 times to check consistency. Custom Python agents: Try the included sample or connect your own agent through an adapter and tool mappings. Platforms: Windows and Linux support, with experimental macOS support. Reports: Inspect individual attacks, evidence, and findings in the Assessment Explorer. It's still an early beta, and I'm looking for around 10–15 developers willing to try it out. I'd especially like to know: Is the setup process straightforward ? Are the test results and evidence actually useful for debugging? What attack scenarios or integrations am I missing? It's intended for authorized, synthetic, nonproduction testing only. Its called FaultFence , more info in comments. Any honest feedback or criticism would be appreciated. Thanks!