init-brain — scaffold a knowledge base

This skill is an ordered checklist: each step opens when the one before it has produced its result. It is the setup-time counterpart of the other skills in this set — they discover a vault's conventions at runtime, and this skill creates the conventions they discover. Scoring a vault that already has content is audit-brain, not this skill.

Whenever a step puts more than one decision to the user, it asks them in the round shape the grill-point skill gives in its step 3.

The component reference is the brain-anatomy skill — read it before step 2.

Steps

1. Confirm the target is a greenfield — inspect the target directory. Effectively empty (no agent instructions file — AGENTS.md or CLAUDE.md — and no content trees) is what this skill builds on. Existing content belongs to audit-brain; say so and stop. Then show the user the plan for this run — its steps by number with a 🛑 on each one that stops for the user, then one line of what this step found and one of what it could not find, labelled in the user's language — and wait for the user's go-ahead, or take the dispatch as the go-ahead when this run was sent to work without the user, as a sub-agent; name each later step by its number as it starts.
2. Interview the variable parts only — the anatomy carries opinionated defaults; ask only what varies per vault, in one round, with the default as each question's recommendation — except the roles, which have none and are asked as a choice between layouts — and let the user override any of them:
   • Which roles/areas of work need their own tree? No default: a vault's roles are the user's own, so offer two or three layouts with what each leads to — a tree per role (teaching / research / admin), or one tree with the subjects at its top — the way grill-point turns "I don't know" into a choice.
   • Where each kind of file lands inside a tree — are imported source files and their digests separate folders (default: sources/ and digests/ per subject) or filed together? The other skills cannot file anything without this answer, so settle it here rather than at first use.
   • Vault language — the language of the generated AGENTS.md, indexes, and templates (default: the language the user is writing in).
   • Assessment — does this vault hold exam papers? If not, skip this and record nothing. If it does, present the default from the assessment-layout skill and let the user override it, then record the answer in AGENTS.md alongside the other destinations, since the assessment skills read it there rather than asking again. Ask for the institution's own assessment vocabulary in the same breath — the cognitive-level scale its forms are printed with, and the paperwork a paper has to produce before it may be used. A report is written on the institution's scale, so a vault that does not carry one makes every run ask again, or worse, borrow a scale from somewhere else. Where the user does not have it to hand, scaffold the slot with a gap marker rather than leaving nothing: an unfilled slot is a question waiting to be answered, an absent one is a question nobody knows to ask.
   • Copyright policy — what may be stored as source files, what is digest-only with citation, and whether openly licensed work (Creative Commons, open access) may be stored with its licence recorded (default: digest-only with citation, and openly licensed work stored with its licence).
   • Privacy red lines — which personal data must never enter the vault. Take concrete categories (default: student names, ID numbers, individual scores, patient records), since a rule written as "no names" reads as a ban on every document that has an author.
   • Staging folder name (default temp/, untracked).
   • Naming convention (default lowercase, hyphen-separated, with issuing org and year).
   • Shell for the check script, PowerShell or bash (default: the one this machine has).
   • Git? If yes, offer a pre-commit hook that runs the check script (default: yes, with the hook, where git is installed).
3. Scaffold every component — walk the brain-anatomy skill top to bottom and create each component's scaffold with the interview answers filled in. Record any component the user declines as declined in the generated AGENTS.md.
4. Verify before finishing — run the project's check script where one exists, then confirm: the vault is discoverable — every operating rule an agent needs (destinations, gates, naming, markers, cadence) is written in AGENTS.md, and every folder is reachable from an index. The bar: the digest skill's discovery step could run here without asking the user anything the interview already answered — walk that step literally, question by question, and treat any question the constitution cannot answer as an unfinished scaffold rather than a detail to settle on first use.