Encode Any Multi-Step Workflow as a Slash Command: Claude Code Skills

TL;DR: Drop a Markdown file into .claude/skills/name/SKILL.md and /name becomes a reusable team command — $ARGUMENTS passes text from the call, frontmatter restricts tools, and disable-model-invocation: true gates side-effect workflows.

Claude Code ships with /code-review, /debug, and a handful of built-ins. Every file you add to .claude/skills/ creates your own /command — committed to git so your whole team gets it, loaded only when called (no CLAUDE.md bloat).

File structure

.claude/
  skills/
    pr-review/
      SKILL.md      →  /pr-review
    changelog/
      SKILL.md      →  /changelog
  commands/         # legacy flat .md files still work
    commit.md       →  /commit

User-level skills go in ~/.claude/skills/ and are available across all your projects.

A complete example

Save this as .claude/skills/review/SKILL.md:

---
description: "Review a specific file or module for correctness and security"
tools: Read, Grep, Glob
disable-model-invocation: true
---

Review the code at $ARGUMENTS for:
1. Logic errors and off-by-one edge cases
2. Security issues (injection, auth bypass, exposed secrets)
3. Missing error handling

For each finding: file path, line number, severity (high/medium/low), one-line description.

Invoke it with:

/review src/auth/login.ts

$ARGUMENTS receives everything after the command name — a file path, a task description, a ticket number, whatever fits the workflow.

Frontmatter options that matter

Field	Effect
description: "..."	Shown in /help and used for Claude's auto-invocation decisions
tools: Read, Bash	Restrict which tools Claude can call inside this skill
disable-model-invocation: true	Only you can trigger it — Claude won't auto-run it

disable-model-invocation: true is the one to know for side-effect workflows like /deploy or /push — you want to be the one pulling that trigger, not Claude deciding the moment is right.

Commit and share

Skills in .claude/ are just files. Commit a useful /pr-review skill with your codebase and every contributor gets the same workflow on clone. No configuration, no onboarding step.

Claude Code skills follow the Agent Skills open standard, so well-formed SKILL.md files work across other compatible AI dev tools too.

Sources:

• Claude Code Skills docs
• Agent Skills standard