SuperAPI AI Directory

setup

Office & productivity

View repository

Updated 10-15 19:54

About this skill

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.