bootstrap

bootstrap — create the brain

Interview the operator and create the minimum brain. Do not ask about scope: the building layer is activated per initiative, not per user.

Step 0 — install the OS if it is missing

Everything below assumes the product is hooked into the brain (.os/core resolves). If it does not — you were installed standalone, e.g. from skills.sh — install it first:

1. Clone github.com/pedroromeroluna/ai-first-os to a fixed local folder outside the brain (the operator will need it to update later).
2. Run <clone>/core/install.sh <brain-path> — idempotent; it hooks the session contract, the resolvers, the subagents and all sixteen skills by symlink.
3. Continue with the interview below.

If .os/core already resolves, skip this step entirely.

The interview

Five questions. One at a time, and do not move on until you have the answer.

1. What is your name?
2. Profile: what you do, what you are on today, what you decide.
3. Voice: how you write when you write well — language, register, words you do not use.
4. How you want to be answered: what you expect from an agent. Length, order, what annoys you.
5. Your organizations: where you work. For the typical case —one person, one employer— that is one. For each one ask what it does and what you do there; the role is written by the system.

Return the answers summarized and wait for approval before writing anything.

What it writes

Build an answers file and run the deterministic part:

.os/core/lib/bootstrap.sh --brain . --answers <file>

Format of the answers file, one key per line:

name: <name>
profile: <one item>
voice: <one item>
reply: <one item>
org: <Name> | <role> | <owner> | <identity file>

profile, voice, reply and org repeat. In org, the owner and the identity file are optional: with no owner the operator is used, and with no identity a placeholder text is left.

The result is operator.md, voice.md, resolver.md, tree.md and one folder per organization. voice.md is the operator's voice — it used to live in a section of operator.md, and from now on it is its own file: one identity, two questions, never the same sentence in both. Nothing else: the inbox, the root's own work (initiatives, backlog, decisions, learnings) and the environment table are born with their first piece of data, like any node.

It runs once per brain. If the root pieces already exist, the script stops instead of rewriting them. Adding an organization to a brain that already exists is new-org's job.

When it finishes

• Show the tree created and what each file answers, in outcomes and not in canonical names.
• If the identity cap warning appears, pass it through as is: the operator chooses the way out.
• If the script reports sin dato: or sin crear:, pass it through whole and ask again for what is missing. A gap nobody names is a gap nobody fills.
• Handoff: if the operator wants to add another organization, the add an organization capability is needed. If it is not available, say so and continue.