Claude Code Setup

A comprehensive, production-ready configuration for Claude Code demonstrating best practices for customization and automation. This is a reference implementation—fork it, steal what you like, adapt it to your workflow.

What's Here

2 Agents: Specialized assistants for specific tasks (evaluator, test runner)
17 Skills: Reusable capabilities for auditing, authoring, workflows, and more
7 Hooks: Automation for validation, formatting, logging, and notifications
Decision guides and references: Help choosing the right component type and naming things consistently

This directory (~/.claude) is the global configuration directory for Claude Code. All customizations here apply across projects unless overridden locally.

Installation

Don't install this. Just steal what you like.

Quick Start

Customize your settings
- Edit settings.json to adjust tool permissions and MCP servers
- Edit CLAUDE.md to document your coding principles and preferences
Create customizations
- Use /create-agent [name] to build specialized agents
- Use /create-skill [name] to create reusable capabilities
- Use /create-command [name] to build quick shortcuts
- Use /create-output-style [name] to define behavior modes
Review the decision guides
- references/decision-matrix.md - Quick component selection
- references/when-to-use-what.md - Detailed scenarios and examples

Directory Structure

Configuration Files

File	Purpose
`settings.json`	Global permissions, MCP servers, cleanup policies, and tool approvals
`CLAUDE.md`	Instructions for Claude when working in this repository
`.gitignore`	Git ignore rules for this configuration directory

Extension Directories (tracked in git)

Directory	Purpose
`agents/`	Specialized AI agents for specific workflows
`skills/`	Reusable capabilities and knowledge domains
`hooks/`	Event-driven automation and validation
`references/`	Shared decision guides and naming conventions

Session Data (not tracked in git)

Directory	Purpose
`projects/`	Per-project metadata and usage tracking
`todos/`	Session-scoped todo lists
`plans/`	Implementation plans from plan mode
`file-history/`	Change tracking for edited files
`session-env/`	Environment snapshots per session
`logs/`	Session and commit history logs
`debug/`	Session debug output
`shell-snapshots/`	Shell environment captures
`statsig/`	Feature flag evaluation cache
`history.jsonl`	Conversation history across sessions

Customizing Your Setup

Creating Agents

When to use: Build specialized assistants for complex tasks requiring specific tools, models, or focused behavior.

/create-agent my-agent

The agent-authoring skill guides you through:

Defining purpose and scope
Selecting model (Sonnet/Haiku/Opus)
Configuring tool restrictions
Writing focus areas and approach

Examples: Read-only analyzers, code generators, domain-specific experts

Creating Skills

When to use: Encapsulate domain knowledge, best practices, or complex workflows that multiple agents/commands might use.

/create-skill my-skill

The skill-authoring skill guides you through:

Defining capability and trigger patterns
Structuring with progressive disclosure
Organizing supporting documentation
Configuring allowed tools

Examples: Best practices, auditing guidelines, deployment procedures

Creating Commands

When to use: Build quick shortcuts for specific workflows or as entry points to skills.

/create-command my-command

The command-authoring skill guides you through:

Designing delegation patterns
Handling arguments and validation
Keeping commands focused and simple
Deciding when to delegate to skills

Examples: User shortcuts, project templates, explicit workflows

Creating Output-Styles

When to use: Define persona modes that change how Claude behaves (tone, verbosity, approach).

/create-output-style my-style

The output-style-authoring skill guides you through:

Defining persona and role
Specifying concrete behaviors
Deciding scope (user vs project)
Setting keep-coding-instructions

*Examples

...

philoserf/claude-code-setup

README