Prime your AI assistant in one command

Kirimana has no web console. Your large language model (LLM) assistant is the interface, so the first thing to do after installing the command line interface (CLI) is teach that assistant everything kiri can do. One command does it:

kiri onboard --agent claude   # writes CLAUDE.md, then run `claude` here

Now open your assistant in that directory and just say what you want to build. It already knows every command, the order to run them in, and the guardrails.

What kiri onboard writes

kiri onboard composes a single, self-contained briefing:

• an operator primer — how Kirimana works (contract-first, the bronze, silver, and gold medallion, plan then apply then verify) and the rules the assistant must respect;
• the full command surface, every command tagged with a side-effect class so the assistant knows what it may run on its own versus what it must hand back to you;
• the build methodology, the canonical sequence from an empty repository to a released warehouse;
• the ready-made skills it can install and run for you;
• your project’s contract digest, once you have a project.

The briefing is generated from the live CLI, so it never drifts from the version you have installed. Re-run it after upgrading kiri.

Choose how your assistant reads it

# Editor agents that auto-load a context file:
kiri onboard --agent claude    # writes CLAUDE.md
kiri onboard --agent cursor    # writes .cursor/rules/kiri.md
kiri onboard --agent codex     # writes AGENTS.md

# Any other assistant (a plain chat):
kiri onboard                   # writes a portable llms.txt to paste in

When you point it at an editor-agent file, kiri onboard writes inside its own marked block, so your own notes in that file are left untouched. An existing file it does not own is never overwritten unless you pass --force.

No project needed

kiri onboard works the moment pip install kiri-cli finishes, before you have created anything. Run it in an empty directory to brief an assistant, then let it walk you through the CLI quickstart. Once you do have a project, re-run it there and the briefing gains that project’s contracts.

Go deeper

For a live connection, so the assistant can read your catalogue, lineage, and compliance status directly rather than from a snapshot, wire up the Model Context Protocol server. The onboarding briefing points at it too.