Lab 07: Agent Red-Team Fuzzer

Series: AI Eval Labs · Phase: Robustness and Security · Depends on: Lab 01

Core scope

• [ ] Inject prompt injections into agent inputs and tool outputs
• [ ] Feed malformed tool schemas
• [ ] Trigger infinite-loop conditions; verify guardrails

Stretch

• [ ] Mutation-based fuzzing with a coverage score

Links

• Code: series/ai-eval/07-red-team-fuzzer
• Demo: coming soon
• Write-up: coming soon on markrox.ai