# Dante

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Faraaz Ahmed](https://cerebralvalley.ai/u/kyokasuigetsu)
- **GitHub:** https://github.com/Ahmed-devL/chaos-hook
- **Demo video:** https://youtu.be/fZN_JRqEoPU
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4758

ChaosHook: AI-powered verification and repair for event-driven backends

Webhook handlers can silently corrupt business state when events arrive out of order, are duplicated, execute concurrently, or are retried after a crash. A successful HTTP response does not guarantee that a customer's balance, subscription, or transaction ledger is correct.

ChaosHook is a local-first developer workbench that combines Meta Muse-powered code repair with deterministic, independently specified verification.

Its Python execution engine tests a vulnerable subscription webhook handler against 24 event permutations, a controlled concurrent race, and two crash/retry schedules using real SQLite transactions. An independent executable business contract compares expected and observed database outcomes, while a React investigation workspace exposes the source code, execution matrix, SQL traces, state differences, and candidate evidence.

Meta Muse Wonder proposes repairs, with Muse Spark 1.3 supporting diagnosis and diagram-based analysis. Candidates are admitted through source-bound verification, authentication-preservation checks, and auditable acceptance gates.

Following an external adversarial audit, I extended ChaosHook with causal checkpoints and counterfactual lifecycle scenarios. The resulting suite executes 62 semantic schedules and 526 checks, rejecting a hardcoded-outcome implementation that fooled the original verifier.

In three live contract-driven Muse Wonder trials, three distinct generated candidates passed both the original 27-check gate and all 526 strict semantic checks. Their retained results were independently reproduced.

ChaosHook also offers a CLI, deterministic evidence exports, mutation testing, and regression coverage. It is a bounded synthetic research prototype rather than being  a claim of exhaustive formal proof or production Stripe integration.

The core principle: AI proposes the repair; independent executable evidence determines whether it passes.

## More from Meta Global AI Developer Hackathon

- [Loreg](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4755)
- [Rikuy.si](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4756)
- [Trade Matrix Labs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4757)
- [Vedit Agrawal](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4759)
- [Jonathan](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4760)
- [Sponge](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4761)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4758. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt