A skill is the highest-level atomic thought in the Rosie architecture. It bundles the discipline (the when and why of a domain) with the operations that implement it (the what and how).

A skill lives in two places:

• .claude/skills/<slug>/SKILL.md — the discipline document, loaded by Claude Code when a task matches the skill's description frontmatter. This is what Claude consults for "should I apply this skill here?" and "what does the discipline say?"
• src/skills/<slug>.md — the runtime view, rendered as /skills/<slug>/. It inlines the SKILL.md, lists the skill's operations as cards, and generates a copy-to-clipboard Claude Code prompt.

A skill aggregates one or more operations. Each operation in turn may aggregate one or more shortcodes (when the operation's contract has multiple invocation patterns, like the changeProposal operation's three modes: Micro-editing, Micro-RAGing, Micro-flagging).

The full architectural relationship across atomic thoughts, skills, operations, and shortcodes is captured in the atomic-thought-architecture diagram concept.