Aegis-arena

Aegis Arena is a CLI-first AI security simulation framework for DeFi teams. We ingest real 2026 exploit incidents from public sources, normalize them into structured datasets, and turn each case into a replayable attacker-vs-defender scenario. Teams can run controlled simulations, compare outcomes across runs, and evaluate response quality with consistent scoring. The project combines a developer CLI, a web control plane, and reproducible simulation packaging so security researchers and protocol builders can train on realistic adversarial behavior instead of static post-mortems.

We built Aegis Arena as a TypeScript Turborepo with a CLI app, a Next.js web app, and modular packages for runtime, ETL, 0G adapters, and simulation orchestration. The ETL pipeline pulls incidents from Rekt News and Solodit, parses and normalizes them into JSONL artifacts, and stores them for replay and analytics. The agent runtime executes attacker, defender, and judge loops with epoch logs, while Kurtosis is used for reproducible isolated environments. We integrated Safe-based session flows and Privy-verified auth for secure browser-assisted CLI login. A notable hack was deriving reliable Solodit metadata from sitemap slug structure when page content was client-hydrated, enabling robust ingestion without brittle browser automation.