Stop Guessing, Start Measuring: A Developer’s Guide to Stress-Testing LLM Agents with Claw-Eval