Follow‑up: how I structure the context tree (Easy AI Context)

Last week I open‑sourced Easy AI Context — a tiny MCP server that gives AI a durable context tree. A few people asked how I structure it, so here’s the shape that’s working for me.

Context tree pattern (example)
projects/<name>/
goals
stack
decisions
status
next
links

Why this layout

• goals + status keep the system anchored
• decisions avoids re‑litigating old choices
• next makes every agent run end with a concrete update

Workflow

1. list_paths → pick the smallest relevant branch
2. get → read only what you need
3. set → write back a short, structured update

This keeps memory curated and durable, not a noisy chat log.

Repo: https://github.com/xiaopai20/easy-ai-context

Curious: do you prefer small curated trees or large auto‑logged memory for long‑term AI work?