Claude's Agent Skills represent an advanced system that enhances its language model capabilities through specialized prompts. Unlike traditional programming models that rely on code execution, Claude operates through prompt expansion. This means that when a skill is invoked, it doesn't run scripts but modifies the conversation context by injecting domain-specific instructions. This approach allows Claude to adapt its responses based on the context provided by these skills, which are essentially templates rather than executable code. The Skills are organized into folders containing instructions and resources, allowing Claude to load them dynamically. Each skill is defined in a markdown file named SKILL.md, which includes frontmatter and detailed instructions. The frontmatter configures how the skill operates, while the content specifies the actions Claude should take. Skills are discovered from various sources, such as user settings and built-in skills, and they allow Claude to perform tasks more effectively by packaging user expertise into reusable resources. When users send requests, Claude evaluates the skill descriptions embedded in the Skill tool's prompt to determine which skill to invoke. This process relies on Claude's language understanding rather than algorithmic intent detection or pattern matching. The interaction is straightforward: Claude modifies the conversation and execution contexts based on the selected skill and continues with enhanced capabilities. Skills guide the workflow rather than executing tasks directly, which distinguishes them from traditional tools in the system.