Learn · Configuration & workflows · CCAR-F task 3.2

Custom commands and skills

Turn a repeated prompt into /review for the whole team, and package occasional procedures as skills that load only when used.

Key ideas

Project commands live in .claude/commands/ and are shared through Git; personal ones in ~/.claude/commands/.
Skills live in .claude/skills/<name>/SKILL.md and can bundle scripts, templates and reference docs.
Only a skill's name and description are in context until it is used (progressive disclosure).
Claude can auto-invoke a skill whose description matches; disable-model-invocation: true makes it user-only.
Frontmatter: description, argument-hint, allowed-tools, model, context: fork (isolated subagent).
Arguments arrive as $ARGUMENTS or $0, $1… (0-based).
CLAUDE.md for every-session rules; a skill for a procedure used sometimes.

Committed as .claude/commands/review.md (or as a skill), it gives the whole team the same /review, with pre-approved read-only tools for the run.

Putting a shared command in ~/.claude/commands/: teammates won't get it.
Putting a long, occasional procedure in CLAUDE.md, where it costs context every session.
Thinking $1 is the first argument: positional arguments are 0-based, so $0 is first.
Forgetting that a skill's description drives whether Claude invokes it automatically.