madbench:madbench-evals

Authors, runs, and debugs madbench evals — bench YAML, checks, red-state testdata, the two controls (check / grade), expectation tuning.

Part of madbench.

4 reference documents, about 1,354 lines.

The skill itself does not contain them. It reads your question and names the one or two worth opening — which is the point, because loading everything below to answer one question would cost more than the answer is worth.

Start here

Document	Covers	Lines
checks-catalog.md	madbench checks catalog	452
debugging.md	Debugging madbench benches	237
runners-and-sandbox.md	Runners (harnesses), permission modes, sandbox	313
schema.md	madbench bench file schema (BenchSpec)	352

Reading these

You do not open these yourself in normal use — /madbench:madbench-evals picks for you. Read them directly when you want the reference rather than the answer, or to check what the advice is actually based on.