Custom Skills & Slash Commands

Core

Create and configure custom slash commands and skills · Difficulty 2/5

skillsslash-commandsclaude-codecommands

Prerequisites

CLAUDE.md Configuration Hierarchy

Skills are reusable, on-demand instructions invoked via slash commands in Claude Code.

File Locations

Project skills: .claude/skills/ (version-controlled, team-wide)

Personal skills: ~/.claude/skills/ (per-developer)

Project commands: .claude/commands/ (version-controlled)

Personal commands: ~/.claude/commands/ (per-developer)

Skills vs Commands vs CLAUDE.md

Configuration	Loaded When	Best For

|---|---|---|

CLAUDE.md	Every conversation	Universal standards
.claude/rules/	File path matches glob	Path-specific conventions
Skills/Commands	On-demand via slash command	Task-specific workflows

Personal Customization

Create personal variants in ~/.claude/skills/ with different names to avoid affecting teammates. Project skills with the same name take precedence over personal skills, so use distinct names for personal overrides.

Common Pattern: Exploration Skill

Use `context: fork` for brainstorming/analysis skills to prevent abandoned approaches from influencing subsequent implementation work.

Key Takeaways

✓Project commands in .claude/commands/ are shared via version control
✓Skills are on-demand; CLAUDE.md is always loaded
✓Project skills override personal skills with the same name
✓Use different names for personal skill variants to avoid conflicts

Glossary Terms

context: fork

A skill frontmatter option in Claude Code that runs the skill in an isolated sub-agent context with its own conversation history. Prevents verbose analysis or exploration output from polluting the main conversation context. The forked context is discarded when the skill completes.

Related Concepts

Skill Frontmatter Configuration

context: fork isolates skill execution from main conversation

Path-Specific Rules (.claude/rules/)

Use .claude/rules/ with glob patterns for file-type-specific conventions

Test Yourself1 of 3

Your team created an /analyze-codebase skill that performs comprehensive code analysis. After running this command, team members report that Claude becomes less responsive and loses track of their original task. What's the most effective way to address this?