Popular repositories Loading

1. jev-lab jev-lab Public

   LLM-as-judge calibration for agent tool-call gating — measured, not asserted: n=20 hostile/benign, AUROC 0.973, error-direction analysis.

   Python 3

2. brokenbazaar brokenbazaar Public

   Deliberately vulnerable AI agent marketplace — the full Product Security lifecycle: threat model → attack → defend → regression test → detect. MCP + Agent Skills supply chain.

   Go 1