Assay: Testing Whether Agent Skills Actually Fire

Field notes

Assay defines YAML case sets containing positive and negative requests, then repeats each case against an agent skill and records the resulting measurements. Its example run covers nine cases and ten attempts each, reports 88 passes and 2 failures across 90 attempts, and gives a 49%–94% confidence interval for one 80% result. The page also treats unknown as a separate verdict and refuses regression comparisons when the runs are not demonstrably comparable.