Claude Code Monorepo

claude-code-monorepo

**Claude Code skills, hooks, subagents, `CLAUDE.md` files, MCP config and memory — the real, sanitized configuration from seven months of shipping a production TypeScript monorepo.**

[](LICENSE) [](https://docs.anthropic.com/en/docs/claude-code) [](SANITIZATION.md)

A near-complete reference export of the Claude Code setup I used to build and ship a 13-service TypeScript platform — roughly 250k lines, ~780 merged PRs, seven months, largely as a single engineer.

This is not a starter template. It is the **real configuration**, sanitized: full skill texts, full hook scripts, the actual memory tree, and one complete epic walkthrough. The company, product, people and infrastructure identifiers have been renamed to consistent fictional ones — see [`SANITIZATION.md`](SANITIZATION.md).

---

The idea

"Loop engineering is replacing yourself as the person who prompts the agent. You design the system that does it instead." — [Addy Osmani](https://addyosmani.com/blog/loop-engineering/)

Two disciplines run through everything here:

• Maker ≠ checker. The agent that does the work never grades it. A separate agent confirms. done is a claim, not a proof.
• Verifiable stop conditions. Every "is it finished?" must be machine-checkable — a command that exits 0, a named RED test that turns GREEN. Never a vibe.

---

Six layers

Layer	Question it answers	Where
Context	What must it always know?	`project/**/CLAUDE.