Agent Skills: A Data-Driven Analysis of Claude Skills for Extending Large Language Model Functionality
Analyzed 40,285 skills to study how Claude skills extend large language model functionality.
Key Findings
Methodology
The study uses a data-driven approach to analyze 40,285 publicly listed Claude skills from a major marketplace. By collecting and analyzing metadata, it reveals trends in skill publication, length distribution, usage patterns, and safety risks.
Key Results
- Skill publication shows short-term burst growth, especially from mid-January to early February 2026, increasing from 2,179 to 40,285 skills.
- Skill content is highly concentrated in software engineering workflows, while information retrieval and content creation account for substantial adoption.
- Most skills remain within typical prompt budgets, but there is a significant supply-demand imbalance and intent-level redundancy.
Significance
The study reveals the potential of Claude skills as an emerging infrastructure layer for agents, providing a quantitative snapshot of skill reuse, standardization, and safety-aware design. These findings help understand the dynamics of the skill market and guide future skill development and application.
Technical Contribution
The paper systematically analyzes the Claude skills ecosystem, revealing patterns in skill publication and usage, as well as potential safety risks. It provides a comprehensive understanding of the skill ecosystem through a data-driven approach.
Novelty
This is the first large-scale analysis of the Claude skills marketplace, revealing trends in skill publication and usage patterns, as well as potential safety risks.
Limitations
- The study is based solely on public marketplace data, which may not cover all skills.
- The safety assessment of skills relies on automated tools, which may lead to misjudgments.
Future Work
Future research could focus on skill standardization and safety design, developing more effective skill discovery and deduplication mechanisms.
AI Executive Summary
Claude skills, as reusable modules, extend the functionality of large language models. By analyzing 40,285 publicly listed skills, the study reveals trends in skill publication, usage patterns, and safety risks. It finds that skill publication often occurs in short bursts, with the highest concentration in software engineering workflows. Additionally, information retrieval and content creation skills have substantial adoption, but there is a significant supply-demand imbalance and intent-level redundancy. Most skills remain within typical prompt budgets, but there are non-trivial safety risks, including skills that enable state-changing actions. The study provides a quantitative snapshot of skill reuse, standardization, and safety-aware design, revealing the potential of Claude skills as an emerging infrastructure layer for agents. Future research could focus on skill standardization and safety design, developing more effective skill discovery and deduplication mechanisms.
Deep Analysis
Background
With the widespread application of large language models, Claude skills, as reusable modules, offer a new way to extend model functionality. Skills define triggering conditions, procedural logic, and tool interactions, rapidly spreading in public marketplaces.
Core Problem
Despite the rapid growth of skill marketplaces, the types of skills available, how users adopt them, and the risks they pose remain unclear. This makes it challenging to understand the dynamics of skill markets and their potential to extend model functionality.
Innovation
This paper is the first to conduct a large-scale data-driven analysis of the Claude skills marketplace, revealing trends in skill publication, usage patterns, and safety risks. It provides a new perspective on understanding the dynamics of skill markets.
Methodology
- �� Collected metadata from 40,285 publicly listed skills. • Analyzed trends in skill publication and length distribution. • Studied usage patterns and supply-demand imbalances. • Assessed safety risks of skills.
Experiments
Data was collected from the skills.sh marketplace, analyzing skill publication dates, installation numbers, and usage platforms. Community attention was tracked using the GitHub API.
Results
Skill publication shows short-term burst growth, especially in software engineering workflows. Information retrieval and content creation skills have substantial adoption, but there is a significant supply-demand imbalance.
Applications
Claude skills can be used in software development, information retrieval, and content creation, helping users automate workflows and increase productivity.
Limitations & Outlook
The study is based solely on public marketplace data, which may not cover all skills. Additionally, the safety assessment of skills relies on automated tools, which may lead to misjudgments.
Plain Language Accessible to non-experts
Imagine you are in a kitchen cooking a meal. Claude skills are like the various tools and ingredients in the kitchen, helping you complete your dish faster and better. Each skill has a specific use, such as chopping, cooking, or seasoning. The skill marketplace is like a supermarket, offering a variety of tools and ingredients for you to choose from. Although there are many options, finding the right tools and ingredients is not easy, especially when they look similar. The study helps us better understand how these tools are used and the potential risks, just like understanding the nutritional value and allergy risks of ingredients.
ELI14 Explained like you're 14
Hey, buddy! Imagine you're playing a super complex game with all sorts of gadgets and skills. Claude skills are like these gadgets, helping you level up faster in the game. Each skill has its own specialty, like fighting monsters, upgrading gear, or solving puzzles. The skill marketplace is like the game store, where you can choose from various gadgets. But sometimes, there are too many gadgets, and picking the right one can be tricky. The study is like a game guide, helping you find the best gadgets and avoid pitfalls!
Glossary
Claude Skills
Claude skills are reusable modules used to extend the functionality of large language models.
In this paper, Claude skills are used to define triggering conditions, procedural logic, and tool interactions.
Data-Driven Analysis
Data-driven analysis is a method that reveals trends and patterns by collecting and analyzing data.
The paper uses data-driven analysis to reveal the dynamics of the skill marketplace.
Skill Marketplace
A skill marketplace is a platform where users can publish and access various Claude skills.
The paper analyzes 40,285 skills from a major skill marketplace.
Safety Risks
Safety risks refer to issues such as privacy breaches or system state changes that skills may cause.
The paper assesses the safety risks of skills and identifies non-trivial risks.
Supply-Demand Imbalance
Supply-demand imbalance refers to the mismatch between the number of skills published and actual user demand.
The paper reveals a significant supply-demand imbalance in the skill marketplace.
Open Questions Unanswered questions from this research
- 1 How can redundancy in the skill marketplace be better removed?
- 2 How can the safety of skills be improved to avoid potential risks?
Applications
Immediate Applications
Software Development Automation
Developers can use Claude skills to automate common development tasks, such as code generation and debugging.
Long-term Vision
Intelligent Assistants
Claude skills can be used to develop more intelligent personal assistants, helping users manage daily tasks.
Abstract
Agent skills extend large language model (LLM) agents with reusable, program-like modules that define triggering conditions, procedural logic, and tool interactions. As these skills proliferate in public marketplaces, it is unclear what types are available, how users adopt them, and what risks they pose. To answer these questions, we conduct a large-scale, data-driven analysis of 40,285 publicly listed skills from a major marketplace. Our results show that skill publication tends to occur in short bursts that track shifts in community attention. We also find that skill content is highly concentrated in software engineering workflows, while information retrieval and content creation account for a substantial share of adoption. Beyond content trends, we uncover a pronounced supply-demand imbalance across categories, and we show that most skills remain within typical prompt budgets despite a heavy-tailed length distribution. Finally, we observe strong ecosystem homogeneity, with widespread intent-level redundancy, and we identify non-trivial safety risks, including skills that enable state-changing or system-level actions. Overall, our findings provide a quantitative snapshot of agent skills as an emerging infrastructure layer for agents and inform future work on skill reuse, standardization, and safety-aware design.