Claude SkillsLesson 6 of 15

How to Use Claude Skills

Invoke skills with slash commands, let Claude pick them up automatically, and use them in both Claude Code and claude.ai.

Two ways to invoke a skill

Skills can be triggered in two ways: call them explicitly with a slash command, or let Claude detect the right skill from your request.

Method	How it works	Example
Explicit slash command	Type `/skill-name` in the Claude input. Claude runs that skill immediately, regardless of context.	/commit
Auto-detection	Claude reads the `description` field of every installed skill and picks the best match when your request fits. No slash needed.	can you format these meeting notes?

To stop auto-detection for a specific skill, add disable-model-invocation: true to its SKILL.md frontmatter. The skill will only run when you type the slash command explicitly.

Using skills in Claude Code

Start a Claude Code session with claude in your terminal. Skills installed in ~/.claude/skills/ (global) or .claude/skills/ (project) are automatically available. Type / to see the list, or type the slash command directly.

claude

# Start Claude Code in your project

$ claude

# See all available skills

> /

Available skills: /commit /code-review /daily-standup

# Run a skill explicitly

> /code-review

→ Running skill: code-review

Reviewing src/auth/token.ts...

Found 2 issues: 1 high, 1 low severity

See full report above.

Skills respect scope: a skill in .claude/skills/ inside your project folder is only available in that project. Skills in ~/.claude/skills/ are available everywhere.

Using skills in claude.ai

On claude.ai, skills are managed from claude.ai/customize/skills. Installed skills appear in the chat input as slash commands.

claude.ai/customize/skills

My Skills+ Add skill

/commit

Stage all changes and write a descriptive commit message

Global

/code-review

Review code for bugs, style issues, and security risks

Global

/daily-standup

Guide a team through a 5-minute standup

Global

claude.ai

/commit

Running skill: /commit

Staged all changes. Here's the commit message:

feat(auth): add token refresh with exponential backoff

Implements retry logic with jitter to avoid thundering herd.
1 file changed, 34 insertions(+), 2 deletions(-)

Combining multiple skills

Skills are independent — you can run several in one session. A common workflow is to run a review skill first, fix issues, then commit.

claude

# 1. Review code before committing

> /code-review

→ Running skill: code-review

1 high: missing input validation in src/api/users.ts:42

# 2. Fix the issue, then commit

> /commit

→ Running skill: commit

fix(api): add input validation to users endpoint

1 file changed, 8 insertions(+)

You can also reference one skill from another via context: fork in SKILL.md — covered in the Skills with Subagents lesson.

Quick reference

Action	How
Run a skill explicitly	/skill-name
See all installed skills	Type / in Claude Code, or visit claude.ai/customize/skills
Let Claude pick the skill	Describe your task — auto-detection uses the description field
Disable auto-detection	Add disable-model-invocation: true to SKILL.md frontmatter
Limit to explicit calls only	Add user-invocable: true (and disable-model-invocation: true)
Hide skill from / menu	Add user-invocable: false to SKILL.md frontmatter
Run skill in subagent	Add context: fork to SKILL.md frontmatter
Check skill is loaded	Ask Claude: 'What skills are available?'

Before you continue

Type /skill-name to run a skill explicitly, or describe your task and let auto-detection pick the right one.
Global skills live in ~/.claude/skills/; project skills live in .claude/skills/.
Use disable-model-invocation: true when a skill should only run on explicit slash commands.
Chain skills in one session — review, fix, then commit is a common sequence.
Next: editing instructions and updating skill behaviour over time.

What's Next

You're using skills confidently. Next: editing instructions and updating skill behaviour over time.

← PREVIOUS

Build Skills from Scratch

How to Edit and Update Skills