Using Skills¶

Skills are folders containing a SKILL.md file with instructions Claude loads dynamically. Unlike prompts, skills are meta-tools that inject domain-specific knowledge into Claude's context when relevant.

How Skills Work¶

Skills use progressive disclosure: Claude sees skill descriptions in its system prompt, decides which are relevant, and only loads the full instructions when needed — preserving context window for actual work.

Invocation Methods¶

Method	Example
Automatic	Claude decides based on the skill description
Explicit	`/skill-name` or "Use the code-review skill"
Disabled auto	Set `disable-model-invocation: true` in frontmatter
Hidden	Set `user-invocable: false` — background knowledge only, not in `/` menu

Official Anthropic Skills¶

These are production-ready and available to paid plans:

Skill	Purpose	Trigger
docx	Word document creation/editing	"Create a Word document..."
pdf	PDF creation, form filling, extraction	"Fill out this PDF form..."
pptx	PowerPoint presentations	"Create a presentation..."
xlsx	Excel spreadsheets	"Create a spreadsheet..."
skill-creator	Meta-skill for building skills	"Help me create a skill for..."

Install via:

/plugin install document-skills@anthropic-agent-skills

For the Sartiq team plugins, see the Marketplace Overview.

Listing Available Skills¶

/skills

This shows all skills from:

Personal skills (~/.claude/skills/)
Project skills (.claude/skills/)
Plugin-provided skills

Skill Storage Locations¶

Location	Scope	Git Status
`~/.claude/skills/`	Personal (all projects)	Not committed
`.claude/skills/`	Project (team-shared)	Committed
Plugin-provided	From installed plugins	Managed by plugin

Team recommendation: Put shared skills in .claude/skills/ and commit to git.

Triggering Skills¶

Automatic Triggering¶

If a skill's description matches your request, Claude loads it automatically:

Review this code for issues

If a code-review skill exists with a matching description, Claude will load and follow it.

Explicit Triggering¶

/code-review

Or:

Use the code-review skill to check the changes in src/auth/router.py

Dynamic Context Injection¶

Skills can inject dynamic context using the `!command` syntax in SKILL.md:

## Current Branch
`!git branch --show-current`

## Recent Changes
`!git log --oneline -5`

Commands prefixed with ! are executed when the skill loads, injecting their output into the skill's context. This lets skills adapt to the current project state.

Custom Commands Are Now Skills¶

The .claude/commands/ directory is deprecated. Custom commands have been merged into the skills system. Migrate existing commands by converting them to SKILL.md format in .claude/skills/.

For the full reference, see official docs.

Debugging Skills¶

If a skill doesn't activate:

Check the description has clear trigger phrases
Ensure no frontmatter syntax errors (YAML)
Verify the skill appears in /skills
Try explicit invocation to confirm it works
Add the skill to CLAUDE.md as a backup reference