With the family

atril works on its own (see the overview). Karajan Code depends on it and uses it as the store for its own commands.

What kj init does

kj init installs Karajan’s kj-* commands into the atril store under .karajan/ (.karajan/commands/), on demand: they are not copied into .claude/commands/ or any other CLI folder, so they do not occupy context until an agent asks for them. Running it twice leaves the store untouched (unchanged).

The /atril bridge

CLIs only accept slash commands they know. atril registers a single one in each CLI, the bridge /atril <name or task>, plus a router rule that tells the agent to ask atril before starting a task. Karajan’s commands, and anything else in the store, are reached through that bridge.

atril sync from kj

When the project already uses rulesync, kj init runs atril sync --root .karajan for you: it writes the bridge and the router rule into .rulesync/ and, if rulesync.generate: true is declared in .karajan/atril.json, generates the files for every CLI. Without rulesync, kj installs the commands, tells you to run rulesync init and then atril sync --root .karajan, and stops there: nothing persistent is written.

What kaRAGan and the HU board add

• kaRAGan indexes the project, so an agent answers with kj rag query instead of guessing what the code does.
• The HU board (kj hu) holds the cards each piece of work belongs to.

atril decides which command or rule reaches the agent; kaRAGan and the board decide what the agent knows and what it is working on.