Markdown Memory

mm — markdown memory & workflow system

A file-based memory and prompt bridge that connects claude.ai (where you think) ↔ Claude Code (where you build) ↔ Obsidian (shared long-term memory).

-lightgrey.svg)

What is this?

When you build with Claude, your work is split across two places: you sketch ideas and plan in claude.ai, then do the real work in Claude Code in your terminal. Those two sides don't share memory, and context evaporates between sessions, between chats, and whenever the context window compacts.

mm closes that gap. It's a set of Claude Agent Skills plus a small PowerShell installer that gives you:

- One shared memory for every project, stored as plain markdown in an Obsidian vault (passport, handoff, session history).
- A prompt bridge so an idea worked out in claude.ai becomes a clean, self-contained task you paste into Claude Code.
- A simple lifecycle — start a project, resume it, save a session, hand off to a new chat — with no copy-pasting your whole history every time.

Everything is markdown and files. No database, no cloud, no lock-in.

How it works

mm spans three environments that share state through files, not magic:

- claude.ai side is a single skill (mm-web-bridge): it challenges your idea, checks fast-moving tech against the live web, and composes the prompt you'll run in Claude Code.
- Claude Code side is the mm dispatcher + 11 mm-* skills that do the work and keep the vault up to date.
- Obsidian is just the folder where the markdown lives. Obsidian indexes it for you; mm reads and writes the files directly.

Quick start

Prerequisites

- Claude Code CLI
- Python 3 (installed on macOS/Windows by default)
- Node.js (v16+ for npx installer)
- An Obsidian vault (any folder works; Obsidian itself is optional but recommended)

NPM / npx…