skill-creator-advance

What problem does it solve?

Help authors reliably create, evaluate, and improve Claude skills by providing a guided, eval-driven development loop that drafts SKILL.md, runs with-skill and baseline evaluations, grades results, detects regressions, and packages distributable .skill artifacts. It reduces manual trial-and-error, surfaces flaky or non-discriminating tests, and improves triggering accuracy via description optimization.

Core Features & Use Cases

• Iterative Eval Loop: Draft → test (with baseline) → grade → improve → repeat, with support for quick vs full eval paths.
• Automated Grading & Benchmarking: Spawn grader/comparator/analyzer subagents, aggregate statistics, and generate benchmark reports and HTML visualizations.
• Description Optimization: Generate trigger evals, run train/test splits, and iterate descriptions using the claude CLI to avoid under-/over-triggering.
• Packaging & Automation: Validation, packaging to .skill, and helper scripts for report generation and benchmark aggregation.
• Use Case: A developer converting a multi-step workflow into a robust skill can draft SKILL.md, supply a few realistic test prompts, run parallel evals, inspect grading, fix issues, and produce a packaged skill for distribution.

Quick Start

Ask: "Help me create a new Claude skill — draft SKILL.md, propose test cases, run evals with baselines, grade results, and suggest improvements until tests pass."