Claude Skills Guide
Turn workflows into slash commands
A Claude skill is a packaged workflow instruction that Claude can execute on demand. Instead of re-explaining how you want something done every session, you encode it once in a SKILL.md file and invoke it as a slash command. This guide covers how to build skills that actually work.
What a skill is
Think of a skill as a detailed standard operating procedure that Claude follows when invoked. It can include step-by-step instructions, decision trees, output formats, quality checks, and anything else that defines how a workflow should be executed.
Once a skill is defined in a plugin and installed, it becomes a slash command: /plugin-name:skill-name. Claude reads the SKILL.md at invocation time and follows it as the authoritative instruction set for that workflow.
SKILL.md format
Every skill starts with a frontmatter block, followed by the skill body.
--- name: my-skill user-invocable: true description: "What this skill does and when to use it" --- ## Step 1: Do this first Instructions for step 1. ## Step 2: Then this Instructions for step 2. ## Step 3: Finish with this Final step instructions.
name
The skill identifier. Combined with the plugin name to form the slash command: /plugin:name.
user-invocable
Set to true to register this skill as a slash command users can invoke directly. Required for /slash-command access.
description
What the skill does. Claude uses this to decide when to proactively suggest or use the skill. Be specific about triggers.
Writing good skill bodies
The skill body is where most people underinvest. Vague instructions produce vague behavior. The best skills are written like instructions for a smart person who has never done this task before.
Be explicit about sequence
Use numbered steps. Claude follows them in order. If the order matters, make it obvious. If a step depends on a previous one, say so.
Define done
Include a completion criteria section. What does a successful execution look like? What should be verified before the skill is considered complete?
Handle edge cases
What should happen if a file does not exist? If a command fails? If input is ambiguous? Skills that handle edge cases gracefully feel reliable. Skills that do not feel broken.
Include quality checks
Add explicit checks at the end: review for common mistakes, validate output format, confirm nothing was skipped. Claude will follow these if you include them.
Practical examples
These are skill types that deliver immediate value:
Stage, commit with conventional commit format, run quality checks, and push to the current branch. Ensures nothing gets committed with secrets, debug logs, or unreviewed changes.
Generate a pull request with a structured description based on the diff. Summarizes what changed, why, and what to test.
Scaffold a new UI component following the project's conventions. Includes the component file, a test file, and an export update.
Review staged changes for common issues before commit — security vulnerabilities, missing error handling, console.logs, and code style.
Need help putting this into practice?
Sprintt helps teams implement AI strategy, build product operating models, and ship faster. Reach out for a consultation.