Claw-SWE-Bench: A Benchmark for Evaluating OpenClaw-style Agent Harnesses on Coding Tasks
CLAW-SWE-BENCH is a new benchmark protocol designed to fairly compare different ways agents are built, called 'harnesses,' like OpenClaw. It standardizes evaluation by fixing everything except the harness itself, allowing researchers to see how much a specific agent design impacts performance under consistent rules for prompt and budget.