ExplainBench: Evaluating Code Explanations from Agents

ExplainBench evaluates whether coding-agent explanations accurately capture the intended behavior of a change and the actual effect of an agent-generated patch.