I Read the Claude Code Docs. All of Them. Here's What I Built.

Claude Code ships with roughly 40 discrete tools, a hook system covering 13 lifecycle events, and an Agent tool that can spawn subagents as flat tool calls. Most people use it as a single-session chat — type a request, get a response, move on. I spent six weeks reading every piece of documentation I could find about those primitives. Not the "getting started" guides — the actual behavior specs. How PreToolUse hooks can return exit code 2 to hard-block a tool call. How CLAUDE.md instructions get loaded into every session and every subagent. How agent markdown files with YAML frontmatter define specialist behaviors. How the Agent tool dispatches subagents with isolated contexts. I wanted to know what happens when you actually compose those primitives. Not by building an external orchestration layer, not by wrapping the API in a custom framework, but by wiring together the pieces Claude Code already exposes. What if hooks weren't just for logging, but for enforcement? What if agents weren