Commands Cheatsheet

Fresh

Skill Locations

# Personal (all projects)
~/.claude/skills/<name>/SKILL.md

# Project (this repo)
.claude/skills/<name>/SKILL.md

# Plugin
<plugin>/skills/<name>/SKILL.md

# Legacy commands (still work)
.claude/commands/<name>.md

Priority Order

enterprise > personal > project (plugin uses namespace, no conflicts)

Frontmatter Fields

---
name: my-skill                    # /slash-command name (lowercase, hyphens, max 64)
description: What it does         # RECOMMENDED - triggers auto-invocation
argument-hint: "[issue-number]"   # Autocomplete hint
disable-model-invocation: true    # Manual-only (default: false)
user-invocable: false             # Hide from / menu (default: true)
allowed-tools: Read, Grep, Glob   # Restrict tools
model: claude-sonnet-4-5-20250929 # Override model
context: fork                     # Run in subagent
agent: Explore                    # Subagent type (Explore, Plan, general-purpose, custom)
hooks: {}                         # Lifecycle hooks
---

String Substitutions

Syntax	Expands To	Example
$ARGUMENTS	All args	/skill foo bar -> foo bar
$ARGUMENTS[0]	First arg	/skill foo bar -> foo
$ARGUMENTS[1]	Second arg	/skill foo bar -> bar
$0	Shorthand for $ARGUMENTS[0]	Same as above
$1	Shorthand for $ARGUMENTS[1]	Same as above
${CLAUDE_SESSION_ID}	Session ID	sess_abc123

Dynamic Context

# Preprocessor syntax - runs before Claude sees content
!`shell command here`

# Example
- PR diff: !`gh pr diff`
- Changed files: !`gh pr diff --name-only`

Permission Rules

# Allow specific skills
Skill(commit)
Skill(review-pr *)    # prefix match

# Deny specific skills
Skill(deploy *)       # prefix match

# Deny all skills
Skill

Common Patterns

Reference Skill (Inline Knowledge)

---
name: api-conventions
description: API design patterns for this codebase
---
When writing API endpoints...

Task Skill (Manual Trigger)

---
name: deploy
description: Deploy to production
disable-model-invocation: true
---
1. Run tests
2. Build
3. Push

Research Skill (Forked)

---
name: deep-research
description: Research a topic thoroughly
context: fork
agent: Explore
---
Research $ARGUMENTS...

Visual Output Skill

---
name: visualize
description: Generate interactive visualization
allowed-tools: Bash(python *)
---
Run the script...

Environment Variables

Variable	Purpose
SLASH_COMMAND_TOOL_CHAR_BUDGET	Override skill description context budget
CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD	Set to 1 to load CLAUDE.md from --add-dir

Useful Commands

# Check what skills are available
"What skills are available?"

# Check context budget
/context

# Invoke with arguments
/my-skill arg1 arg2

# Add external skill directories
claude --add-dir /path/to/skills