open-source · since 2026
MiDojo
Asago
A red-teaming framework for AI agents that intercepts tools and data sources, injects controlled attacks, and grades both task utility and resistance to prompt injection using simulated or sandboxed environments.
AI safetyagentsred teamingprompt injectionMCPOpenShell
desk notes
Verified 2026-09-29 via Asago's live site, GitHub API and README. Active through 2026-09-29; supports MCP, PI and OGX agent paths plus dict and OpenShell environment backends. No tagged release yet.