Popular repositories Loading

1. trifecta-guard trifecta-guard Public

   Runtime security middleware that blocks lethal prompt-injection capability chains in AI agents.

   Python

2. abstain-bench abstain-bench Public

   Calibration-first LLM benchmark for confident-error and abstention scoring.

   Python

3. passk-reliability passk-reliability Public

   Domain-agnostic pass^k reliability harness for tool-using agents.

   Python

4. mast-doctor mast-doctor Public

   MAST failure-mode classifier and dashboard for multi-agent traces.

   Python

5. rewardhack-probe rewardhack-probe Public

   Reward-hacking toy environments and inference-time probe battery.

   Python

6. mcp-sentinel mcp-sentinel Public

   MCP tool-poisoning and rug-pull watchdog for MCP servers.

   Python