Agent Evaluation

Identifies critical patterns for LLM agent evaluation but lacks actionable instructions

• +Identifies key anti-patterns like single-run testing and output string matching
• +Highlights critical sharp edges like benchmark-production gaps and data leakage
• -Provides no executable instructions or procedures for conducting evaluations
• -Declares filesystem write permissions without apparent justification