audit-evals
Audit evals
Find the cells of a regression suite that no longer earn their run, from the cell history alone, and move to the lower tier only the ones the user approves. A cell is one case judged by one check. The upper tier runs whenever the suite runs; the lower tier runs only when the model the product or skill runs on changes, and a lower-tier cell that fails there goes back up as a regression. Demotion keeps every cell: nothing here deletes a case or a check.
This skill requires llm-evals, which runs the paid verification described
below. Before doing any work, confirm it is installed. When it is missing, stop
and report it with its install command,
npx skills@latest add toy-crane/skills --skill llm-evals, using the project's
package manager runner where it pins one.