cocottetech/@platform/codebase/@features/ai-copilot/docs/D-onboarding.brief.md

D — Onboarding + persona-seed interview

Goal

First-run experience: get Quinn from cold-install to a working ai-copilot in under 5 minutes, with a seeded personas row that ai-copilot + content-* specialists can lean on from day one.

Designer skim

• Headline UX: 5-minute first-run. SSO device-link → conversational persona-seed interview (8–12 Qs via voice or text) → ai-copilot summarizes + Quinn approves → handoff to chat-home.
• Flows (3): D1 SSO device-link · D2 persona-seed interview · D3 first-run empty state.
• States (7): splash · device-link in progress · device-link error · mid-interview turn · persona summary review · persona save error · onboarding complete.
• Pair-with: persona-seed-interview.screen.md.
• Blocking Qs: OPEN-DECISIONS.md → D-Q1 voice-vs-text default, D-Q2 interruptible TTS.

Constraints

• Auth via SSO device-link against quinn.sso (federated, P0); will migrate to platform.sso later — UI should be auth-provider-agnostic.
• Persona-seed interview is voice + text: Quinn can speak answers (TTS prompts; STT capture) OR type. Mixing is fine.
• Persona is written via PATCH /api/v1/personas/:user_id once the user record exists.
• Reuse lilith-messenger-ios/Core/Auth/{AuthManager,SSOAuthenticator,KeychainService}.swift — port unchanged.

Flows

D1 — SSO device-link

1. Splash with brand mark (CocotteAI).
2. Tap "Sign in" → opens device-link QR / code displayed on iPhone.
3. Quinn scans on a trusted device OR enters the code in quinn.sso's portal.
4. Device link confirmed → JWT issued → Keychain storage → home.
5. If user has no personas row → push to D2.

D2 — Persona-seed interview

A guided conversation, ~8–12 questions, surfaced as a chat thread with ai-copilot. NOT a form.

Question shapes (designer to refine wording):

• "What's your brand voice in one sentence?" (free text, ~140 chars)
• "Pick 3–5 words your audience would say about you." (chip select with custom-add)
• "Who's your audience? Walk me through a typical fan." (free text / voice)
• "What surfaces do we plan for? OF / X / IG / TikTok / Tryst / TS4Rent / others?" (multi-select)
• For each selected surface: "Different voice here than your default? (yes/no — if yes, how?)"
• "What's off-limits? Topics, kinks, places I should never bring up." (free text, multiline)
• "What's a typical week like? When do you shoot? When do you post?" (free text / voice)
• "Anything I should know to not embarrass you?" (catch-all)

After: ai-copilot summarizes the persona back to Quinn and asks for sign-off. On approval → write personas row.

D3 — First-run empty state (after persona seed)

• "Welcome — strategist is drafting your first 14 days. Back in ~30 seconds."
• Progress indicator while strategist runs (post-P0 — for P0 this state is "We're ready when you upload your first shoot or DM us a question").

States to design

• Splash (signed-out).
• Device-link in progress (waiting on QR scan).
• Device-link error (timeout, wrong code).
• Mid-interview turn (question + voice/text input options).
• Persona summary review.
• Persona save error.
• Onboarding complete → handoff to chat home.

In-the-wild copy

Splash CTA (hearth, dialed warmer — first impression):

Welcome. Set the kitchen up.

Device-link in progress (working):

Scan the code on a trusted device. Cocotte will know it's you.

Mid-interview question, voice or text (hearth):

Two more, then she'll let you go. What's a typical week? When do you shoot, when do you post?

Persona summary review (working — Quinn is deciding):

Here's what Cocotte heard. Voice: warm, dry, no marketing. Off-limits: real name, ex, Vegas. Surfaces: OF, X, Tryst, TS4Rent. Save and start, or rework anything?

Onboarding complete, handoff (hearth):

She's ready. Drafts will be in the drawer by morning.

Out of scope

• Multi-account switching (Quinn's only operator P0).
• Reset-and-redo persona (lives in persona-editor drawer, brief B5).

Open questions

• Voice-first or text-first as the default input mode?
• Should the interview accept Quinn typing while ai-copilot is still speaking the next question (interruptible TTS)?
• Skip-to-end affordance, or force completion?