AgentBehaviorBench
AgentBehaviorBench (ABB): A benchmark for evaluating AI agents’ ability to test other agents and detect behavioral defects.
- 22
- Python
- MIT
- 3 d ago
AgentBehaviorBench (ABB): A benchmark for evaluating AI agents’ ability to test other agents and detect behavioral defects.
CRAB (Clean Rebuild of Anything's Behavior): rebuild software from behavior, not source code. An MCP-powered toolkit for black-box observation, clean-room reimplementation, and behavioral verification.
Executable behavioral contracts and hidden-evaluation harness for coding agents across nine software domains.
Evaluation-driven RAG system using RAGAS, retrieval metrics, behavioral stress testing, holdout validation, diagnostics and reliability analysis.