Claude Code Memory Stack

The Claude Code Memory Stack

[](https://github.com/icatchstyle/Claude-Code-Memory-Stack/actions/workflows/ci.yml) [](LICENSE) [](https://www.python.org/downloads/) [](CONTRIBUTING.md)

A complete, opinionated, **runnable** template for giving Claude Code a persistent memory, a skill library, and an MCP layer — derived from a setup that has been in daily production use, then stripped of everything company-specific.

This is a working system rather than a reading list: a vault skeleton you copy, a `CLAUDE.md` you fill in, hooks that fire on every turn, skills that trigger when you want them, and a reference MCP server that runs. Everything is designed to be adopted incrementally — you can stop after Level 1 and still be better off.

**Contents** — [The problem](#the-problem-this-solves) · [What you get](#what-you-get) · [Quick start](#quick-start) · [The five ideas](#the-five-ideas-that-make-this-work) · [Adoption levels](#levels-of-adoption) · [Docs](docs/) · [Contributing](CONTRIBUTING.md)

---

The problem this solves

Claude Code is excellent at reasoning over the code in front of it and **amnesiac about everything else**. Every session it re-learns which database profile is safe to query, which branch to fork from, that the container needs a restart after the mount goes stale, that the build breaks if you edit the compiled CSS instead of the source.

You can paste that into a prompt. You will paste it again tomorrow.

The fix is not a longer `CLAUDE.md`. The fix is a **layered architecture** where each kind of knowledge lives in the layer that can actually deliver it at the right moment:

Layer	Holds	Loaded
`C