Skill Creator: author and evaluate skills

Turn a skill’s intent into SKILL.md, then improve triggering and performance with test prompts and evaluation.

Skills

Use when creating, editing, or evaluating an agent skill. Clarify the intended behavior, trigger conditions, output format, and boundaries, then draft SKILL.md. For objectively checkable tasks, prepare test prompts and iterate using qualitative review and quantitative measures. If a draft already exists, focus on evaluation and improvement. Confirm file locations and scope before writing; benchmark runs can consume model calls, so agree on scale and cost first. Subjective creative skills can use reader feedback instead of forced metrics.

Permissions

Prerequisites

A Claude environment that supports Anthropic Skills and a basic description of the intended behavior, trigger, and output; running evaluations also requires an available Claude runtime and evaluation tools.

Permissions

May create or edit SKILL.md, write test prompts, and run skill evaluations within the user’s authorization; confirm file scope and evaluation cost before broad edits or benchmark runs.

Limitations

Evaluation depends on the model, prompt set, and runtime; subjective writing skills may need qualitative review, and a single run is not a general guarantee.