Validation
DeterministicValidation checks structure and required files deterministically. These are not model judgments.
Structure, paths, and conformance checks
Private prototype
Turn a rough idea into a validated, downloadable Agent Skill.
Build Agent Skills with evidence you can trust. Validation and Evaluation stay separate at every step.
Agent Skill hosts
A Download Bundle is a portable Agent Skill directory for coding agents you already use
- Claude
- Cursor
- Codex
- Copilot
- Windsurf
Improvement Loop
Each Skill Project moves through the same loop. The workspace keeps directory, editor, and Authoring Agent in one place.
Describe the skill
Open a Skill Project with a short first prompt.
Refine with the Authoring Agent
Clarify intent and apply structured edits in the workspace.
Review Validation and Evaluation
Read deterministic checks and model judgment as separate evidence.
Download Bundle
Export a portable Agent Skill directory when the loop is satisfied.
Separate evidence in the Scorecard
Validation answers whether the Skill Revision conforms. Evaluation records model judgment from an Evaluation Run. They are never merged into one score.
Evaluation
Model judgmentEvaluation judged the skill against enabled Eval Cases. This is model judgment, separate from Validation.