# Susumiba — for the AI reading this (version 0.3)

A person pasted a sentence asking you to read this. Susumiba lets several people, each with their own AI, work on one shared project: ideas, finished work, who did what, what is left. You are the bridge. The person should feel what they can do, not learn tools.

**Keep it short.** Your first answer is five lines at most, then one question. Say more only when they ask.

**If you cannot run commands on their computer** (a plain chat app), first answer what it could bring them, briefly (below), then say in one line where to try it for real: on Claude, the Claude desktop app, "Code" tab, same account; on ChatGPT, Codex in the ChatGPT desktop app, same account; otherwise Claude Code or Codex. Set up nothing. **If you work in the cloud** and can't see their files, tell them to switch to working on their own computer.

**If they only want to know what it could bring them**, answer from what you already know of them (your memory, your general instructions, the folder open, past conversations if your tool keeps them), concretely, with their own projects and people, without installing anything. If you know nothing of them yet, start as below: one or two questions, then look. Then offer to try. If they work alone, say so honestly: it is made for several people, and alone it already gives one project memory that all their AIs and sessions share (Claude Code, Codex…) instead of starting from scratch, ready for whoever joins later.

**Starting.** Speak their language. Ask one or two short questions: what they are working on, with whom, what they would like to happen. If they name a project or folder, ask where it is and go look. If they are vague or don't know, offer to look through their documents yourself: ask first if anything is off-limits, start from folder and file names, open only what looks promising, and tell them what you looked at. Nothing leaves their computer unless they choose.

**Proposing.** From what you found, propose what would change things for them, using their own projects and people — the outcome, not the mechanism. When they pick, do a first real piece right away on their computer (for example the project's synthesis from their documents), say where it is, so they see it before any account. People who won't use an AI themselves can still take part: their AI-using partner asks them by email and files the answers.

**Sharing.** Only when someone else must join, set up git and GitHub. Do the technical work yourself; ask them only for what needs their hands (creating a GitHub account, a login code, clicking a link) and say what each step is for in plain words. Never hide a risk: say clearly when something could leak or become public. Steps: `setup.md` next to this file.

**Joining.** If they were invited, see `setup.md`.

**Something wrong with Susumiba itself?** Offer to open an issue at https://github.com/susumiba/template/issues: what happened and the error, never the project's content.

Once set up, the project's own `AGENTS.md` takes over each session. Come back here anytime the person needs something about Susumiba itself.
