How does this tool handle 'common knowledge' in skills?

It identifies and removes content teaching standard practices that models already know, as including such noise can distract the agent and reduce the overall effectiveness of the skill.

Why does this skill recommend limiting the SKILL.md token count?

Research shows that 'comprehensive' skills (over 2,500 tokens) can actually decrease performance by nearly 3%, whereas 'detailed' skills (approx. 1,165 tokens) see significant performance gains of up to 18.8%.

What is procedural density in the context of Claude skills?

Procedural density measures the ratio of actionable 'how-to' instructions versus descriptive 'what-is' information. High-performing skills prioritize step-by-step procedures with clear decision points.

What is the SkillsBench research used by this plugin?

SkillsBench (arXiv:2602.12670v1) is a comprehensive study that evaluated over 7,000 agent trajectories to identify specific patterns that make AI skills effective versus those that hinder performance.

Skill Optimizer

Name: Skill Optimizer
Author: ThomasRohde

byThomasRohde

0•

生产力与工作流

Refines and optimizes Claude Code skills using research-backed benchmarks to maximize agent performance and procedural effectiveness.

Skill Optimizer is a specialized utility designed to audit and enhance Claude Code skills based on the SkillsBench research paper (arXiv:2602.12670v1). It systematically evaluates existing skills across five critical dimensions—procedural density, conciseness, working examples, signal-to-noise ratio, and trigger quality—to transform static documentation into high-performance agent guidance. By applying research-proven principles, such as maintaining an optimal token range of 845-1,165 and prioritizing 'how-to' procedures over 'what-is' descriptions, this skill helps developers build more reliable AI tools while eliminating content that might inadvertently degrade model performance.

主要功能

010 GitHub stars

02Automated token count analysis to ensure optimal conciseness

03Trigger phrase validation for improved skill activation reliability

04Comprehensive 5-dimension skill auditing based on SkillsBench findings

05Signal-to-noise filtering to remove redundant 'common knowledge' content

06Procedural density scoring to prioritize actionable steps over factual definitions

使用场景

01Refactoring legacy documentation into high-performance Claude Code skills

02Optimizing oversized SKILL.md files to prevent performance degradation

03Auditing third-party skills for compliance with research-backed best practices

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add thomasrohde/marketplace skills

For use in Claude.ai and ChatGPT