What problem does it solve? AI features, agent pipelines, and prompts are hard to verify objectively, so teams ship changes without knowing whether quality improved or regressed. This Skill provides a structured method to define success criteria before implementation and measure pass rates after. ## Core Features & Use Cases - Capability Evals: Define objectively verifiable success criteria for new AI features before writing any code. - Regression Evals: Track existing behavior as PASS/FAIL checks to detect breakage from changes. - Grader Selection: Choose between code-based, model-based, or human-based grading depending on output determinism. - Use Case: When tuning a prompt for an agent pipeline, define a capability eval with three verifiable criteria, implement the change, then have a separate grading context score pass@3 against the 90% target before reporting results. ## Quick Start Define a capability eval with three objective success criteria for my new prompt, then grade the results in a separate context.
