Claude Code//.claude//skills

- Reusable instruction packages invoked with `/name` in a session: each skill is a directory inside `.claude/skills/` containing a `SKILL.md` file.

Reusable instruction packages invoked with /name in a session: each skill is a directory inside .claude/skills/ containing a SKILL.md file.

SKILL.md has YAML frontmatter + markdown instructions.

Name:: The slash command name (e.g. commit → /commit)

Description:: What it does. Claude also uses this to decide whether to auto-load the skill when relevant.

Disable-model-invocation:: If true, only the user can trigger it (not Claude on its own)

Argument-hint:: Shown to the user as placeholder (e.g. "[filename]")

Allowed-tools:: Restrict which tools Claude can use while executing this skill.

The body is a prompt: can reference $ARGUMENTS and $0, $1, $2 for positional args.

Skills are not scripts: they tell Claude what to do with existing tools, not deterministic execution.

Legacy path .claude/commands/<name>.md also works but lacks frontmatter features.