The system that gets better every time you use it

After each piece of work, run /retrospective. Sub-agents scan your conversation for corrections, redone work, missing steps, and what worked well.

They propose updates to your skill files and new memories. You review, approve, and the next session is better.

/retrospective              # analyze current session
/retrospective video       # focus on video skill

Full observability. Full control of what gets written to your memory.

Claude Code stores memories in a hidden project folder. This setup brings them into your Obsidian vault with a reverse symlink.

Create a Memories.base to see all memories in a filterable table. Types: feedback, project, user, reference. Sort by last updated.

You can read, edit, and delete any memory file. No black box.

CLAUDE.md is your foundation. Auto-memory learns from every conversation. Auto-dream consolidates overnight.

The retrospective skill is the manual loop on top. You run it after a piece of work. It proposes skill improvements and memory updates from the conversation.

Memory Observability lets you see everything in one place. The system maintains itself. You just do your work.