Claude Code Blueprint

Claude Code Blueprint

A production-grade Claude Code configuration: skills, agents, hooks, rules, and permissions working as a system, not as isolated features.

Companion repo for the article: *"I Turned Claude Code Into an Operating System. Here's the Blueprint."*

Architecture

Each layer reinforces the others. The anti-hallucination protocol works because Context7 is an accessible MCP server. Security hooks work because static permissions define a perimeter. Multi-agent brainstorm works because agents have dedicated tools and models.

What's Included

What's NOT Included

This is a curated subset, not a mirror of a production setup. The full system has 32 skills, more agents, and exact security regex patterns. The hooks bash-guard and write-guard are sanitized: they show what categories of threats to block without exposing exact regex. Adapt them to your threat model.

Quick Start

Directory Structure

The 5-Phase Pattern (AAPEV)

Every skill follows the same cognitive discipline:

This prevents the model from jumping to Phase 4 (code) without understanding (Phase 1), researching (Phase 2), and planning (Phase 3).

Build Your Own

Adding a Skill

Create ~/.claude/skills/your-skill/SKILL.md:

Adding an Agent

Create ~/.claude/agents/your-agent.md:

Adding a Hook

1. Create the script in ~/.claude/hooks/scripts/your-hook.sh
2. Make it executable: chmod +x your-hook.sh
3. Register it in ~/.claude/settings.json under the appropriate lifecycle event

Claude Code supports 17 lifecycle events: SessionStart, UserPromptSubmit, PreToolUse, PermissionRequest, PostToolUse, PostToolUseFailure, Notification, SubagentStart, SubagentStop, Stop, TeammateIdle, TaskCompleted, ConfigChange, WorktreeCreate, WorktreeRemove, PreCompact, SessionEnd. This blueprint covers 9 of…