setup

# Setup — make this AIOS yours A guided interview that turns the blueprint into a personalized AIOS. Read `aios-intake.md` first; if it's filled, confirm and scaffold; if not, run the interview conversationally, **one question at a time, with a recommended answer** so the user can confirm/correct. ## Interview (capture each answer into `aios-intake.md` as you go) 1. **Who are you** — role, company, what you own. (→ `context/about-me.md`, `CLAUDE.md` header) 2. **Voice samples** — ask the user to **paste 1–2 real, unedited** things they've written. *Refuse typed-fresh prose* (it contaminates the sample). (→ `references/voice.md`) 3. **90-day priorities** — dated, concrete. Push back on vague. (→ `context/priorities.md`) 4. **Business** — what it sells, to whom, the model. (→ `context/about-business.md`, Knowledge base) 5. **The stack** — where revenue/customers/comms/docs/tasks live. (→ `connections.md`) 6. **The team** — roster + who can absorb delegated work. (→ `context/about-team.md`) 7. **The constraint / pain** — what eats the week; what to hand off first. 8. **Guardrails** — the hard rules + sensitive-data line + any compliance regime (GDPR/HIPAA/SOC2/...); what may be sent/acted-on vs only drafted. (→ `CLAUDE.md` Guardrails) ## Scaffold (after the interview) Fill every `{{placeholder}}` in `CLAUDE.md` (name, persona, knowledge base, voice summary, connections, **guardrails — customize rule #1 and #8 to the domain**). Write `context/*`, `references/voice.md`, and the `connections.md` rows. Back up any existing versions to `archives/` first. ## Introduce the engine (one screen) Point the user at: `/structure` (frame fuzzy asks) · `/triage` · `/daily` · `/weekly` (Cadence) · `/wiki` (second brain: `raw/` → `wiki/`) · `/graph` · `/graph-query` · `/graph-ingest` (knowledge graph) · `/workflow` (dynamic workflows) · `/experiment` (autoresearch loop) · `/audit` + `/level-up` (weekly). Then suggest the first prompt: *"what should I focus on this week?"* ## Optional: knowledge graph (graphify) Offer to set up the knowledge graph so the user can *see* and query their code + second brain: run `python tools/graphify_setup.py install` (installs graphify and wires the Claude Code & Codex skills). Then `/graph` builds it. Skippable and additive — graphs **code + the committed `wiki/` only**, never `raw/`. ## Rules - Voice paste cannot be skipped or typed fresh. - One-shot scaffold after the interview; the user iterates by editing `aios-intake.md` and re-running. - Customize the Guardrails block to the user's domain — don't ship the generic defaults unchanged for a regulated business.