What Are Skills?
A skill is a named command (e.g., /commit, /review-pr, /deploy) that, when invoked, expands into a full prompt that tells Claude exactly what to do. Think of them as saved expert instructions.
For example, a /commit skill might tell Claude to:
- Check git status
- Review staged changes
- Generate a descriptive commit message
- Create the commit
Opening Skills Settings
- Open Settings (
Cmd+,) - Select "Skills" from the left sidebar
- Choose scope: Global (all projects) or a specific Project
Viewing Skills
The Skills view is a read-only display of all discovered skills. Each skill shows:
- Skill name — The slash command name with
/prefix (e.g.,/commit) - Description — What the skill does
- Model — Which Claude model it uses (if specified)
- Invocation mode — Whether it's user-only, auto-only, or both
- Raw content — Expand to see the full skill file contents
An "Open folder" button lets you jump to the skills directory in your file manager.
Global vs Project Skills
- Global skills are available in all projects and sessions
- Project skills are only available when working in that specific project
Creating Custom Skills
Skills are defined as SKILL.md files inside named directories:
- Global skills:
~/.claude/skills/my-skill/SKILL.md - Project skills:
your-project/.claude/skills/my-skill/SKILL.md
The directory name becomes the command name (e.g., commit/SKILL.md creates the /commit command).
Skill File Structure
---
description: Generate a descriptive git commit message
user-invocable: true
---
Review the staged changes using `git diff --staged` and create a commit with a clear, conventional commit message. Follow the project's commit conventions if a CONTRIBUTING.md exists.Frontmatter Options
| Field | Description |
|---|---|
description | What the skill does (shown in the skills list) |
model | Which Claude model to use |
user-invocable | Whether users can trigger it with /command |
disable-model-invocation | Prevent Claude from auto-invoking |
agent | Reference to a custom agent |
allowed-tools | Restrict which tools the skill can use |
argument-hint | Hint shown when the user types the command |
The body is the prompt sent to Claude when the skill is invoked.
Using Skills
In the chat input, type / followed by the skill name:
/commitClaude receives the skill's prompt and executes the instructions. You can also add extra context after the skill name:
/commit focus on the authentication changesPlugins
The Skills view also shows installed plugins — third-party extensions from the Claude Code marketplace. Each plugin shows its name, version, description, and a link to its homepage.
MCP Servers
MCP (Model Context Protocol) servers extend Claude's capabilities by giving it access to external tools and data sources. Pandev lets you configure MCP servers both globally and per-project.
Agents
Agents are custom agent definitions that Claude can use for specialized tasks. They define agent types with specific tools, instructions, and capabilities.