I am building a basic project‑memory system that uses Obsidian as the storage layer and Claude Code as the AI coding assistant.
I store each project's notes in separate Obsidian folders, tag them with descriptive keywords, and maintain a JSON/YAML mapping file that links project identifiers to those folders and keyword sets.
When a new coding session starts, I query the mapping, compute similarity between candidate notes and the session context, then present the top‑N matches to myself for confirmation before injecting them into Claude Code’s prompt.
I log which notes are used in each session so the similarity scoring can be refined over time.