A skill is a Markdown file that gets injected into the AI's context so it behaves consistently across sessions. It is the persistent memory the tool doesn't have by default.
Large language models forget everything between conversations. Skills give them a reliable reference sheet — a way to answer TanStack Query vs useEffect, the test folder, the casing — once.
Skills live as files in the repository — a plain Markdown document, often with YAML frontmatter.
Teams share them through Git. Source code that affects how the AI behaves — review them in PRs.
Modern implementations support loading skills only when they're relevant, saving tokens and attention.
Third, skills can bundle more than text — supporting scripts, reference documents, and templates the AI can execute or read on demand.
They all point at the same underlying idea: persistent, reusable context separate from the conversation itself.