cowork-memory-kit

A persistent memory system for Claude Code and Cowork. Gives Claude a real long-term memory across sessions — curated topic files for what matters, a full chat archive for everything else, and skills to save, load, and search the lot.

Why I built this

I was sick of re-explaining who I am, what I'm working on, and what we'd already decided every time I started a new Claude session. Out-of-the-box memory tools either dump everything into one bloated file or forget the moment you close the tab. I wanted something structured: a small set of curated files Claude maintains itself, plus a searchable archive of every conversation, plus simple commands to drive the whole thing.

Stack

• Markdown for all memory files (human-readable, hand-editable, diffable)
• Three Claude Code skills: save, memory-load, memory-search
• A claude-memory/ scaffold that drops into ~/Documents and grows over time

How it works

• /save at the end of a conversation — Claude reviews the session, updates the curated memory files (profile, projects, knowledge, people, changelog) with anything worth keeping, and writes the full raw transcript to the chat archive
• /load at the start of a new conversation — Claude reads every memory file and catches up on who you are and what's in flight
• /search anytime — searches across both the curated files and the full chat archive to find something specific from a past conversation
• Over time Claude creates new topic-specific files (e.g. recipes.md, health.md, career.md) when a thread deserves its own home, so the system grows with you instead of being a single dumping ground

What's in the box

• save.skill, memory-load.skill, memory-search.skill — double-click to install in Cowork
• claude-memory/ — empty scaffolding folder with the canonical vault structure:
  • .claude/CLAUDE.md — vault organization rules Claude reads at the start of every session (edit the communication preferences section to match your voice)
  • memory/ — core memory files (profile.md, projects.md, knowledge.md, people.md, changelog.md) plus any custom topic files you add over time
  • chats/ — raw conversation archive
  • deliverables/ — where every file Claude creates for you lives, organized by project
  • assets/, data/, tools/, workspaces/ — standard folders for images, datasets, scripts, and full project workspaces
  • _templates/ — starter templates for new memory files, deliverables, and chat archives
  • _index.md — top-level map of content, updated as new files are created
• claude-memory/skill-source/ — editable source for the three skills if you want to customize the behavior

Setup

1. Double-click each .skill file to install
2. Move claude-memory/ into ~/Documents/ (or wherever you keep your Cowork folders)
3. Open claude-memory/.claude/CLAUDE.md and edit the Communication Preferences section to match how you want Claude to talk to you
4. Fill in claude-memory/memory/profile.md with your name, email, and anything else you want Claude to know about you
5. Type /load at the start of your next Cowork session — Claude will read everything and be caught up

Demo

This is the actual memory system I run every day. The repo only ships the scaffolding — no personal memory content.