woostack-eval

Evaluate one skill package against frozen behavior and trigger corpora. Produce comparative, receipt-backed evidence without changing the implementation under evaluation. This command is an evidence runner, not a model-provider client, editor, merge gate, or automatic fix loop.

The exact data shapes are in the evaluation schemas. Isolation, baseline resolution, preparation, dispatch, and completion follow the runner contract. Read only those directly linked references when detail is needed; do not recursively load unrelated skills or routing guidance.

Invocation

/woostack-eval <skill-path> [--behavior | --triggers | --all]
  [--runs <1..10>]
  [--baseline-ref <git-ref> | --baseline-path <skill-dir>]