AI agents are being deployed rapidly without security testing. Most teams have no way to test their agents against real-world attacks like prompt injection, data leakage, and tool abuse. Sentra solves this for AI engineers and security teams who need to validate agent security before going to production.
A full-stack platform that scans AI agents against 13 attack scenarios mapped to 6 OWASP LLM Top 10 categories. Users paste an agent URL or use the built-in demo agent, and Sentra runs simulated attacks (prompt injection, jailbreaks, token smuggling, tool loop exploits, etc.). It outputs a security score (0–100) with per-OWASP-category breakdown, PDF/CSV/JSON reports, and scan comparison to track regressions.
The 13 unique attack scenarios each needed different payloads and validation logic — OpenCode helped me iterate quickly by generating attack templates and debugging WebSocket real-time streaming. The async subprocess management for Nikto integration required multiple rounds of OpenCode-assisted refactoring.
OpenCode taught me how to structure a complex FastAPI + React app with proper async patterns. I used OpenCode's plan mode extensively to design the OWASP scoring algorithm before writing code, which saved hours of rework. The AI helped me understand WebSocket broadcasting patterns I'd never used before.
Add more OWASP scenarios, a GitHub Action for CI/CD scanning, and multi-agent comparison — all with OpenCode's help to speed up development.
Automated quick review. Add a live URL, screenshots, and a demo video to boost your score.