Why Hjarni alongside CLAUDE.md
CLAUDE.md is the right tool for short, in-repo rules. It is the wrong tool for a growing pile of decisions, runbooks, and incident notes. Hjarni gives Claude Code the larger, searchable layer alongside CLAUDE.md, not instead of it.
Claude Code Stop hooks can write every session to Hjarni once you wire the script (see the hooks doc for the exact setup). After that the loop runs itself: work, ship, the note writes itself, tomorrow's session reads it. No 'I'll write that up later' that never happens.
The same Hjarni knowledge base is reachable from Claude Code, Claude.ai, ChatGPT, and Cursor. The notes you write in the terminal are the notes your assistant reads in the browser. One source of truth, every interface.