Simon Willison argues that Claude Skills could be more important than the Model Context Protocol (MCP), a bold claim that highlights the elegance and potential of this simple yet powerful approach to extending AI model capabilities.
Skills are fundamentally folders containing instructions, scripts, and resources that Claude can load contextually when relevant to a task. This simple definition conceals an important conceptual innovation: instead of building complex infrastructure, Skills rely on the model's ability to read and interpret instructions structured in Markdown, optionally accompanied by supporting documents and scripts.
The technical implementation of Skills is remarkably minimalist. A Skill can be as simple as a single Markdown file containing detailed instructions on how to accomplish a specialized task. Skills are designed to be token-efficient: brief explanations are loaded at the start of a session, allowing Claude to understand when and how to use each Skill without consuming an excessive number of tokens on each interaction.
This efficiency contrasts sharply with heavier approaches. Skills depend on a coding environment with filesystem access, but this dependency is their only infrastructure requirement. This simplicity greatly facilitates sharing and distribution: a Skill can be shared as a single file or a small compressed folder, making the creation of a Skills ecosystem extremely accessible.
Willison identifies several particularly promising practical applications. In the field of data journalism, Skills could automate complex data analysis and fact-checking workflows. Creating specialized agents for specific tasks becomes trivial: instead of building a complex agent architecture, one can simply provide Claude with an appropriate Skill. Extending the model's capabilities happens in an organic and iterative way.
The comparative advantages over MCP are substantial. Skills are inherently more flexible and lightweight, without requiring servers or complex communication protocols. Crucially, they are not limited to Claude: the concept of providing structured instructions can work with different models, making Skills potentially portable across platforms. Token overhead is minimal, and creation and iteration are fast and simple.
Willison predicts a "Cambrian explosion" of Skills, evoking the period of biological evolution characterized by rapid diversification of life forms. He anticipates that the simplicity and accessibility of Skills will lead to widespread adoption and creative implementations across a multitude of domains that we are only beginning to imagine.
The article emphasizes the elegance of Skills' simplicity. In a technology ecosystem often prone to excessive complexity, Skills represent a refreshing return to first principles: clear instructions, relevant context, and direct execution. This minimalist approach could paradoxically prove more transformative than more sophisticated but more complex solutions like MCP.
In conclusion, Willison suggests that we may be underestimating the importance of Skills. Their combination of simplicity, efficiency, and power makes them an ideal mechanism for AI task specialization without complex infrastructure, opening the door to an era of decentralized and accessible innovation in the development of specialized AI agents.