Document Claude skill requirements and produce SKILL.md with evaluation plans.
npx skills add https://github.com/skaisser/blueprint --skill skill-creator-skaisser
Please help me install this Agent Skill. Skill: skill-creator Source: https://github.com/skaisser/blueprint/tree/main/skills/skill-creator Command: npx skills add https://github.com/skaisser/blueprint --skill skill-creator-skaisser
Skills for Claude require a repeatable lifecycle to design, implement, evaluate, and improve them. This unit provides a structured framework to create new skills from scratch, iterate them based on quantitative evals, and optimize triggering and performance over time.
Provide a complete skill in SKILL.md following this unit's structure and optionally bundle scripts, references, and assets for on-demand use.
High-intent search queries and answers about installing and using this skill.
To create a Claude skill, document your requirements and goals, then generate a SKILL.md file with frontmatter and body content that encodes the workflow, required resources, and tooling references for your specific use case.
Evaluation-driven iteration uses an evals.json file and dedicated workspaces to run test prompts against baselines, quantify performance improvements, and guide refinements to your skill over time.
Establish a repeatable benchmarking workflow by defining an evaluation plan with test prompts, expected outputs, and baselines, then iterate using workspace results to measure and optimize skill performance.
Yes, you can optionally bundle supplementary resources by organizing custom scripts, reference documents, and assets into dedicated directories alongside your core SKILL.md file for on-demand use.
Before drafting an evaluation plan, you need to identify and document the skill requirements, including clearly defined goals, intended use cases, and specific success criteria to measure against.
Avoid using an automated skill lifecycle framework if your project lacks defined success criteria or cannot support repeatable evaluation workflows, as quantitative iteration relies on consistent baseline measurements.
maltemdDesign evaluation frameworks with metrics and pass/fail thresholds for LLM-based agents.
pynbj1001Create, iterate, and benchmark SKILL.md skill definitions with eval harnesses.
ReidondRecord AI infrastructure changes in a structured changelog format.
TugoukezhangAnalyze and refine prompts for clarity, structure, and specificity.
