Eve library

OSHUN V1 — Master Design Brief for Claude Design

Guide

Revision 1

On this page

How to use this document. This is a self-contained brief for Claude Design, Anthropic's prompt-to-prototype tool that generates production-ready HTML/CSS/JS prototypes. Every section is written so you can paste it directly into Claude Design as either a full project brief (Part A → Part H) or a per-surface prompt (Part E + relevant cross-cutting sections from Part F). The "Quick-paste prompts" in Part I are pre-assembled for one-shot screen generation.

Status: Authored 2026-05-07. Synthesises canonical OSHUN V1 inputs: docs/releases/v1/design/brand-system.md, docs/releases/v1/design/customer-visual-thesis.md, docs/releases/v1/scope/imagery-illustration-texture-ambient-media-rules.md, docs/releases/v1/design/ux-principles.md, docs/releases/v1/specs/personas.md, docs/releases/v1/specs/product-brief.md, docs/releases/v1/scope/mvp-scope.md, docs/releases/v1/design/motion-language.md, docs/releases/v1/design/ergonomics-requirements.md, plus the design-token foundation in libs/oshun/design-tokens/.


Part A — How Claude Design Reads This Brief#

Claude Design generates interactive HTML/CSS/JavaScript prototypes (not static images) inside an iterative chat canvas at claude.ai/design. Best results come from briefs that have:

  1. Concrete tokens — specific hex codes, font families, pixel values, not adjectives like "elegant blue."
  2. Named references — "calm like Headspace, editorial like Are.na, dense like Linear" anchor the output without prescribing pixels.
  3. One dominant goal per generation — Claude Design works best per-screen or per-flow. Generate the global system once, then prompt one surface at a time referring back to it.
  4. Stated anti-goals — explicit "do not do X" constrains the search space and prevents drift toward generic SaaS aesthetics.
  5. Edge states named — loading, empty, error, offline, partial-failure, reduced-motion, dense-content, and accessibility presets must be enumerated or Claude Design will skip them.

Recommended workflow.

  1. Open a new project in Claude Design. Paste Part B (Foundation) as the system primer. Confirm Claude Design has internalised tokens by asking it to render the colour palette and type scale.
  2. Paste Part C (Personas) and Part D (Product Structure) so it understands the "one shell, five rooms" model.
  3. For each surface in Part E, start a new chat in the same project and paste the section header plus its body. Reference the foundation by name ("Use the OSHUN aqua-500 from the foundation").
  4. Use Part I (Quick-paste prompts) for surfaces you want shipped fast.
  5. Export each finished surface as standalone HTML or a Claude Code handoff bundle.

Part B — Foundation: Brand, Tokens, Visual Thesis#

B.1 One-sentence thesis#

OSHUN customer product should feel like a quiet observatory for inner life: calm and premium at first glance, minimal in chrome, embodied in posture and motion, and unmistakably non-generic in how each domain adds atmosphere without breaking one shell.

Short canonical form (use as governance label):

Calm, premium shells with embodied wayfinding, minimal chrome, and unmistakably non-generic domain atmosphere.

B.2 Brand character#

  • Calm, precise, and trustworthy. Not loud. Not bouncy. Not playful in the consumer-fintech sense.
  • Premium but not ornamental. Richness comes from material contrast, composed typography, and motion cadence — never from glassmorphism, glossy metallics, or excessive blur.
  • Technical confidence with human warmth. A research instrument that still feels emotionally aware. Think: a high-end Leica or a Muji notebook, not a startup landing page.

B.3 Reference posture (use as design anchors)#

Aspect Anchor reference Why
Calm pacing, daily-practice posture Headspace, Calm Emotional regulation, low-friction entry
Editorial restraint, quiet luxury Are.na, Robin Sloan's Subcast, FT Edit Composed typography, wide reading rhythm
Dense functional clarity Linear, Notion Calendar, Things 3 Information legibility under restraint
Embodied wonder Dark Sky (legacy), NASA APOD Wonder without sci-fi clutter
Trust-and-evidence Stripe Press, NYT Times Insider Premium rigor without enterprise severity

These are posture references only. Do not imitate their branding, illustration style, or domain-specific tropes (especially do not imitate wellness-app blob aesthetics).

B.4 Colour system#

B.4.1 Core brand palette#

text
Token            Hex        Purpose
ink-950          #040B16    App chrome, dark anchors
ink-900          #0B1728    Elevated dark surfaces
ink-700          #22334D    Dividers on dark
fog-300          #C8D5E8    Secondary text on dark
fog-100          #EAF0F8    Light-mode backgrounds
white            #FFFFFF    Light surfaces and contrast text
aqua-500         #20C9D8    Primary OSHUN accent (CTA, focus)
amber-500        #F4A524    Highlight / warning emphasis
success-500      #2AB673    Success / positive state
danger-500       #E24D4D    Error / destructive state

Primary shell gradient (nocturnal canvas):

  • Start #08152A → Mid #0A2036 → End #040B16

B.4.2 Domain accent palette ("light in the room")#

text
Domain    Token             Hex        Symbol motifs
Tara      domain-tara       #2BC6B4    lotus, breath, wave
Arete     domain-arete      #3AC77B    spark, target, path
Veritas   domain-veritas    #27A3F3    shield, check, document
Nyx       domain-nyx        #7D7CFF    moon, star, telescope
Nisaba    domain-nisaba     #C88B2C    quill, manuscript, margin

Rules. Domain accent tints local rails, focus areas, progress, and emphasis. Never full-screen domain-tinting of shell canvases or navigation. Domain glows are additive overlays at 10–18% opacity — never colour replacements. Domain identity is "light in the room," not paint on every surface.

B.4.3 Semantic roles#

text
surface/base       canvas background
surface/raised     cards, sheets
surface/overlay    modals, top layers
text/primary       default body and titles
text/secondary     metadata, supporting copy
action/primary     main CTA (aqua-500)
status/{success|warning|error|info}   feedback messaging

B.4.4 Themes#

  • Default mode is dark ("nocturnal, luminous, border-first, calm"). Light mode exists for daytime utility surfaces and reading.
  • High-contrast variants required for both dark and light modes (WCAG contrast ≥ 7:1 for body text in HC modes).
  • Colour is never the sole state indicator — pair with icon or text.

B.5 Typography#

text
Family            Stack
Display/headline  "Space Grotesk", "Manrope", "Segoe UI", sans-serif
Body/UI           "Manrope", "Avenir Next", "Segoe UI", sans-serif
Mono/data         "IBM Plex Mono", "SFMono-Regular", "Menlo", monospace

Type scale (px / line-height / weight):

text
display-xl   48 / 56 / 600    Hero titles
display-lg   40 / 48 / 600    Feature headers
title-xl     32 / 40 / 600    Screen titles
title-lg     28 / 36 / 600    Section titles
title-md     24 / 32 / 600    Card titles
body-lg      18 / 28 / 500    Lead paragraph
body-md      16 / 24 / 500    Default body
body-sm      14 / 20 / 500    Secondary text
label-md     13 / 18 / 600    Buttons, chips
label-sm     12 / 16 / 600    Metadata labels
mono-sm      12 / 18 / 500    Technical values

Reading-frame standards (Nisaba, long-form): 62-character measure, 1.6 line-height, 24px paragraph spacing, drop-cap optional. See docs/releases/v1/design/readability-standards.md.

Typography is the main luxury signal. Do not replace it with chrome.

B.6 Spacing, radius, elevation#

4-pt spacing scale:

text
space-1=4   space-2=8    space-3=12   space-4=16   space-5=20
space-6=24  space-8=32   space-10=40  space-12=48  space-16=64

Radius scale:

text
radius-sm=8    cards within dense lists
radius-md=12   default card / button
radius-lg=16   sheets / large surfaces
radius-xl=24   bottom sheets / hero panels
radius-pill=999  chips, pills, avatar rings

Elevation (prefer borders + spacing over shadow):

text
Level 0  Flat surface, no shadow
Level 1  Card — 1px border at 8% white, shadow 0 1px 2px rgba(0,0,0,0.4)
Level 2  Sheet/modal — stronger shadow, controlled blur
Level 3  Transient overlay (popover/toast) — strongest, brief lifespan

Domain glow halos (when used at all): radial gradient at 12–18% domain accent over a 200–400px diameter, no hard edges.

B.7 Iconography#

  • Primary set: Phosphor (or geometric stroke equivalent — Lucide acceptable for prototypes).
  • Stroke width: 1.75px at 24px base.
  • Sizes: 16 (inline), 20 (dense rows), 24 (default interactive), 28 (prominent nav).
  • Tinting: structural nav icons stay neutral. Domain symbols may tint by domain accent.

B.8 Motion language#

Duration tokens (ms):

text
instant=0    fastest=75   faster=100   fast=150     normal=200
medium=300   slow=400     slower=500   slowest=700

Easing tokens:

text
linear                 linear                          continuous loops
easeIn                 cubic-bezier(0.4,0,1,1)        exits / acceleration
easeOut                cubic-bezier(0,0,0.2,1)        entrances / deceleration
easeInOut              cubic-bezier(0.4,0,0.2,1)      state changes
emphasizedDecelerate   cubic-bezier(0.05,0.7,0.1,1)   priority entrances
emphasizedAccelerate   cubic-bezier(0.3,0,0.8,0.15)   priority exits

Default recipes:

  • Tap feedback: 100–150ms, easeOut, subtle scale or elevation shift.
  • Card entrance: 200–300ms, easeOut, translateY(8–16) + opacity.
  • Modal/bottom sheet: 300–400ms, easeOut in / easeIn out.
  • Route transition: 300–500ms, easeInOut, single direction (don't double up fade + slide).
  • Loading loops: 1000ms max, linear, must stop the moment content arrives.

Sequencing. Container first → primary content → secondary metadata. Stagger tokens: tight 30ms, normal 50ms, relaxed 75ms, loose 100ms. Max 8 elements per stagger sequence.

Reduced-motion. Respect prefers-reduced-motion. Replace movement with instant state changes; preserve cues via opacity/colour. Disable list staggers, ambient loops, and parallax. Route transitions become ≤100ms fade or instant.

B.9 Imagery, illustration, texture, ambient media#

Doctrine. Every hero needs one real anchor. Decorative gradients + particles do not count. If the surface still reads as a generic SaaS app after hiding the media, the media is too weak.

Approved anchor types: photography, editorial illustration, product-integrated media scene, controlled ambient-motion canvas.

Per-domain imagery direction:

text
Tara      Embodied calm, natural light, breath, still water, grounded ritual.
          Avoid pastel-blob wellness clichés.
Veritas   Documentary, source, evidence-adjacent imagery.
          Avoid panic-newsroom chaos, warning-red overload, sterile dashboards.
Nyx       Expansive astronomical, observational tools, precise nocturnal
          atmosphere. Avoid neon sci-fi, HUD clutter, generic space wallpaper.
Arete     Disciplined real-world motion, lived practice environments.
          Avoid trophy/podium gamification, hyper-athletic clichés.
Nisaba    Manuscript, margin, annotation, study-desk materiality.
          Avoid museum-kitsch or fantasy-parchment cosplay.

Texture rules. One dominant texture treatment per section max. Should be legible only after a second look. Acceptable: film grain, paper tooth, star haze, soft radial atmosphere, edge vignetting.

Ambient-media rules. Slow and low-amplitude only. Silent by default. Must ship a still-frame fallback. Reduced-motion freezes immediately. Never auto-play audio. Never use motion to hide weak composition.

B.10 Anti-goals (off-thesis, banned)#

  • Generic SaaS card-grid homepages.
  • Wellness-app clichés (pastel blobs, inspirational filler copy, lotus icons used as primary nav).
  • Luxury-signalling through glassmorphism, excessive blur, or metallic gradient gimmicks.
  • Per-domain visual fragmentation that makes OSHUN feel like five disjoint apps.
  • Multiple competing primary actions on the same viewport.
  • Decorative motion that competes with reading or orientation.
  • Pseudo-spiritual visual language that weakens trust, rigor, or seriousness.
  • Centred hero above a wall of cards.
  • Gradient-only hero with token-rich but visually generic content.
  • Punitive streak logic, guilt-based copy, badge proliferation.
  • Top-corner placement of session-critical actions on mobile.

Part C — Personas and Jobs-to-be-Done#

OSHUN V1 is designed for four primary personas. Every surface should be walkthrough-tested against at least two of them.

C.1 P1-A — The Daily Balancer (24–44, mobile-first, multiple short sessions)#

  • Core intent: reduce stress fast, sustain wellbeing routines.
  • Primary domains: Tara, Arete.
  • Functional JTBD: "When my day is chaotic, help me start a grounding session fast."
  • Emotional JTBD: "Help me feel in control, not behind."
  • Design implications: one-tap quick action on Home, default to short high-confidence routines, simple motivating progress.
  • Success signals: session start <30s of app open; 4+ weekly quick sessions.

C.2 P1-B — The Truth Seeker (20–50, mobile by day + web in deep reading)#

  • Core intent: understand what is true, what is uncertain, why.
  • Primary domain: Veritas (with Tara/Arete decompression).
  • Functional JTBD: "When information is noisy, give me verified, source-grounded summaries quickly."
  • Emotional JTBD: "Help me feel informed without feeling manipulated."
  • Design implications: confidence and evidence visible on every Veritas card, concise summary before deep-dive, fast save/follow.
  • Success signals: high drill-down completion; repeat save/follow.

C.3 P1-C — The Curious Observer (16–40, evening, often outdoors,#

intermittent connectivity)

  • Core intent: turn sky curiosity into regular discovery.
  • Primary domain: Nyx (with occasional Veritas, Tara).
  • Functional JTBD: "When I look up at the sky, help me know what I'm seeing and what's happening tonight."
  • Emotional JTBD: "Help me feel wonder and competence at the same time."
  • Design implications: night-safe contrast, offline fallback, obvious sky exploration entry, no expert-knowledge assumption.
  • Success signals: frequent nightly highlights; star-map exploration completion.

C.4 P1-D — The Intentional Builder (23–48, morning planning + evening#

reflection)

  • Core intent: structured daily action toward long-term growth.
  • Primary domain: Arete (plus Tara for regulation).
  • Functional JTBD: "When I set growth goals, help me convert intention into daily actions and visible progress."
  • Emotional JTBD: "Help me feel momentum, not guilt."
  • Design implications: ≤60s daily check-in, recovery-after-miss patterns, compact streak signals near primary actions.
  • Success signals: high daily check-in completion; recovery-after-miss rate.

C.5 P2 personas (post-MVP design influence)#

  • P2-E Routine Integrator: combines 2–4 domains weekly. Drives cross-domain continuity surfaces.
  • P2-F Accessibility-First User: uses assistive tech by default. Drives semantic rigor and presets.

Part D — Product Structure: One Shell, Five Rooms#

D.1 The shell#

OSHUN ships as three product surfaces sharing one shell:

  1. Mobile app (apps/oshun/mobile) — React Native + Expo, iOS + Android.
  2. Web/PWA (apps/oshun/web) — Next.js, installable PWA with offline shell.
  3. Admin web + admin mobile (apps/oshun/admin, apps/oshun/admin-mobile) — operator surfaces for review, incidents, governance.

D.2 Customer top-level information architecture#

Every customer surface lives under one of these five top-level destinations:

text
Home       Personalised greeting, time-of-day context, continuity strip,
           domain entry cards with continue-state, cross-domain quick actions,
           recommendations with "why this" labels.

Explore    Curated and searchable cross-domain content, blended discovery
           lanes, domain filters, recent and suggested queries.

Activity   Cross-domain timeline of recent progress, streaks, retry/recover
           controls for failed items, study/notebook continuity threads.

Library    Unified saved/bookmarked items across all five domains, collections,
           resume lanes, research-to-practice journeys.

Profile    Account, settings, privacy, accessibility, notifications, theme,
           consent, billing, data rights, connected domains.

Plus three persistent overlays:

  • Universal search (cmd-K on web, search button on mobile) — federates across all five domains with continuity strip + dense results + preview lane.
  • Notifications centre — per-domain delivery controls, read states, priority filters.
  • Assistant panel — global overlay with memory disclosure, continuity journeys, domain handoffs.

D.3 The five domains ("rooms")#

Each domain is a room in the OSHUN home, not a separate app. Domain identity appears through accent, content posture, imagery, and tone — never through replacing shell grammar.

text
Tara      Meditation and mindfulness. Softest, most centring posture.
          Anchor: warmth and breath over gadgetry.
          The default emotional anchor for the broader shell.

Arete     Goals, habits, journaling, accountability coaching.
          Anchor: forward momentum, discipline, clarity.
          Energy comes from crisp hierarchy and motion discipline,
          not gamification.

Veritas   Verified intelligence, fact-checking, evidence-grounded reading.
          Anchor: calm rigor, not newsroom anxiety or enterprise severity.
          Evidence sharpens confidence without cooling the shell.

Nyx       Astronomy, sky events, cosmic perspective.
          Anchor: wonder, scale, perspective.
          Most visually atmospheric domain; still shares shell restraint.

Nisaba    Long-form scholarly reading, lexicon, concept graph, notebook.
          Anchor: scholarship, continuity, reverence for text.
          Editorial and durable; ornament never interferes with measure.

D.4 Two-taps-to-value contract#

For every P1 persona, the primary domain action must be reachable in ≤2 taps from Home:

  • Tara → tap Home → tap "Resume meditation" (Primary Zone, thumb-reachable).
  • Arete → tap Home → tap "Daily check-in" (Primary Zone).
  • Veritas → tap Home → tap top trending claim card.
  • Nyx → tap Home → tap "Tonight's sky."
  • Nisaba → tap Home → tap "Continue reading" rail card.

D.5 Ergonomics contract (mobile)#

  • Thumb zones (portrait):
    • Primary Zone: lower 55% of screen height, hand-side 65% of width.
    • Stretch Zone: middle 30% of height, opposite-side lower region.
    • Hard Zone: top 15% of height, opposite-side upper corner.
  • Tap targets: ≥44×44pt iOS / 48×48dp Android / 44×44 CSS px web; ≥48–56 for primary actions; ≥8px between targets, ≥16px for destructive.
  • Cross-hand neutrality. Core actions usable for both left- and right-hand grip.
  • Top bar reserved for low-frequency contextual actions. Never put repeated session-critical actions in top corners.
  • Domain switcher and assistant launcher must be reachable without grip shift from Home.

Part E — Per-Surface Briefs#

Each section below is a paste-ready surface brief. Header → body. When prompting Claude Design, paste the foundation summary first, then the section.

E.1 Public web — Welcome / marketing landing#

Goal. First impression of OSHUN as a quiet observatory for inner life that opens into five rooms. Convert curious visitor to download/sign-up.

Audience. Cold visitor referred from press, social, or word-of-mouth. May have heard "AI super-app for inner life" but not seen the product.

Hero anchor (mandatory). One dominant real composition — not a gradient-only hero. Suggested treatments (pick one):

  1. Full-bleed nocturnal sky photograph (Nyx-anchored) with a constrained text column over a calm tonal area; subtle controlled ambient motion (slow parallax of stars at low amplitude).
  2. Split composition — left: editorial photograph of a reading desk or morning ritual; right: composed text + CTA.
  3. Immersive ambient field — slow generative visual (e.g., aurora curve moving at 0.02px/frame) anchored by clear text and one CTA.

Layout sections (in order):

  1. Hero with one anchor (above) and a single CTA: "Open OSHUN" (aqua-500). Kicker copy ≤14 words. Headline using display-xl.
  2. The five rooms strip — five quiet cards across (mobile: vertical stack), one per domain. Each card has: domain accent edge (top or left), icon, name, single-line promise, no gimmick. Do not tint the entire card.
  3. One-paragraph thesis with display-lg opener and body-lg body.
  4. Three "moments" panels — typographic feature panels, each with one image anchor, illustrating a moment of use (morning, midday, evening). Not feature lists.
  5. Trust strip — quiet citation row (press, advisors), no logos cloud.
  6. Footer — ink-950 background, fog-300 text, three columns max.

Components to render: Hero with media + text column; five domain-room cards; ambient-field background module with reduced-motion fallback; quiet press strip; minimal footer.

Anti-goals: card-grid above the fold, gradient-only hero, marketing-site cliché ("trusted by"-logo cloud, gradient blob illustrations, animated number counters).

Responsive breakpoints: 360, 768, 1024, 1280, 1440. The hero must work as still frame at 360.

Accessibility: WCAG AA contrast on all text over media (use overlay darkening when needed). Keyboard tab path traverses CTA → five-room strip → trust strip → footer. Reduced-motion freezes ambient field.


E.2 Onboarding flow — sign-up, sign-in, preferences#

Goal. Get a new user from "I just installed" to "first useful session" in under three minutes. Calm, never sales-y.

Audience. Someone with a domain motivation (any of the four P1 personas) who needs to be oriented to "five rooms in one home."

Flow (5 screens max):

  1. Welcome / value primer — single sentence ("A calm place for inner life, organised in five rooms"), one anchor image, two CTAs: "Continue with email" / "Continue with Apple/Google."
  2. Account / sign-in — minimal form, no marketing copy. Forgot-password inline, magic-link path supported.
  3. Domain preference selector — five domain cards with accent edges; user taps any combination ("which rooms feel useful right now?"). Skippable.
  4. First-action picker — one of: "Take a 3-minute breath" (Tara), "See tonight's sky" (Nyx), "Pick today's claim to verify" (Veritas), "Set one habit" (Arete), "Choose a passage to read" (Nisaba). The chosen one becomes Home's continuity card.
  5. Permissions — push notifications, motion (Nyx), location (Nyx) framed in plain language. All deferrable.

Components: auth form, domain selection grid, single-question step cards with skip, system permission primer rows.

Tone of microcopy. Calm, concrete, no exclamation marks. "Let's start" not "Welcome aboard, friend!" Every sentence ≤16 words.

Edge states: invalid email, network failure with retry, returning-user detection (skip onboarding).


E.3 Customer shell — Home (web/PWA)#

Goal. Within one viewport, the user understands time-of-day, sees their continuity card, can reach their primary domain action in one tap, and senses the other four rooms without distraction.

Audience. Primary daily P1 personas (any of the four). 80% of sessions land here.

Layout (desktop, 1440 baseline):

text
┌──────────────────────────────────────────────────────────────────────────┐
│  Top bar: OSHUN wordmark · search · notifications · avatar   (height 56) │
├────────────┬─────────────────────────────────────────────────────────────┤
│            │                                                             │
│  Sidebar   │  Greeting kicker ("Good morning, Elikem") (label-md)        │
│  (240px)   │  Anchor headline (display-lg)                               │
│  Home      │  Subline: time-context (body-md, fog-300)                   │
│  Explore   │                                                             │
│  Activity  │  ┌── Continuity card ──────────────────────────────────┐    │
│  Library   │  │  "Continue: 7-min Calm Drift (Tara)"                │    │
│  Profile   │  │  Resume button (aqua-500, primary)                  │    │
│            │  │  Dismiss · Switch                                   │    │
│  ────      │  └────────────────────────────────────────────────────┘    │
│  Tara ⋅    │                                                             │
│  Arete ⋅   │  Cross-domain quick actions row (4 chips):                  │
│  Veritas ⋅ │  · 3-min breath  · Today's claim  · Tonight's sky           │
│  Nyx ⋅     │  · One habit                                                │
│  Nisaba ⋅  │                                                             │
│            │  Five-room strip:                                            │
│            │  ┌───────┬───────┬───────┬───────┬───────┐                  │
│            │  │ Tara  │ Arete │Verit. │ Nyx   │Nisaba │                  │
│            │  │ teal  │ green │ blue  │purple │ amber │ (accent edge)    │
│            │  │ card  │ card  │ card  │ card  │ card  │                  │
│            │  │ +cont │ +chk  │ +clm  │ +sky  │ +psg  │ (continue-state) │
│            │  └───────┴───────┴───────┴───────┴───────┘                  │
│            │                                                             │
│            │  Recommendations rail (3 cards) with "why this" labels      │
│            │                                                             │
└────────────┴─────────────────────────────────────────────────────────────┘

Continuity card. Surfaces the user's most likely next action across domains. Primary CTA is aqua-500, single button. Secondary actions (Dismiss, Switch) are text/secondary.

Five-room strip. Each card:

  • Top edge accent stripe in domain colour (4px).
  • Domain icon (24px, neutral or domain-tinted as configured).
  • Domain name (title-md).
  • Continue-state line (body-sm, text/secondary).
  • One inline action (e.g., "Resume," "Open").

Recommendations. Each card carries a "why this" chip explaining the recommendation in ≤6 words ("Because you're in evening mode," "Saved 3 articles like this").

Empty / first-time state. Continuity card replaced by "First moments" card pointing to the chosen onboarding action; recommendations replaced by domain primer carousel.

Partial-failure state. Any domain that fails to hydrate shows a quiet "Domain unavailable" inline note inside its card; shell remains rendered.

Anti-goals. No metric chips dashboard. No second sidebar. No animated hero numbers. No "growth"-style trending lists.

Mobile (360–428): sidebar collapses to bottom-tab nav; sidebar's domain launcher becomes the Domain Switcher overlay (see E.18).


E.4 Customer shell — Home (mobile)#

Goal. Same as E.3 but optimised for one-handed thumb-zone use, short sessions, and motion. 60% of OSHUN sessions are mobile.

Layout (414px baseline, portrait):

text
┌──────────────────────────────────┐
│  Time chip · avatar     (top)    │ Hard zone — read-only metadata
├──────────────────────────────────┤
│  Greeting (label-md, fog-300)    │
│  Anchor headline (title-xl)      │ Stretch zone — orientation
├──────────────────────────────────┤
│                                  │
│  Continuity card (full-width)    │
│  Resume CTA (aqua-500, ≥56px)    │ Primary zone — action
│                                  │
├──────────────────────────────────┤
│  Quick-action chip row           │
│  (horizontal scroll, 4 chips)    │ Primary zone — frequent actions
├──────────────────────────────────┤
│  Five-room strip                 │
│  (horizontal scroll, snap)       │ Stretch zone — exploration
├──────────────────────────────────┤
│  Recommendations (vertical)      │
│  Card 1 (full bleed)             │ Stretch / scroll
│  Card 2                          │
├──────────────────────────────────┤
│  [Home][Expl][Act][Lib][Prof]    │ Primary zone — bottom nav
│                                  │ + persistent FAB: Domain Switcher
└──────────────────────────────────┘

Bottom nav. Five-tab IA. Active tab: aqua-500 indicator + label. Tap target ≥48dp.

Domain Switcher FAB. Anchored bottom-left (left-thumb default; settings allow mirroring). Tap opens the Domain Switcher bottom sheet (E.18).

Assistant launcher. Anchored bottom-right. Tap opens Assistant Panel (E.19).

Greeting copy. Time-aware ("Good morning," "Quiet evening"), no name exclamation. Subdued.

Reduced-motion. Five-room strip becomes vertical static stack; recommendations don't animate in.


E.5 Customer shell — Explore#

Goal. A blended cross-domain discovery surface that respects the user's intent but invites surprise. Not a content firehose.

Audience. Users in exploratory mood; lower frequency than Home.

Layout sections (top to bottom):

  1. Search field at top (large, label-md placeholder "Search across OSHUN…"). Cmd-K on web. Always-visible.
  2. Domain filter row — five chips + "All." Toggle, not radio.
  3. Lanes (4–6, stacked):
    • "Today across OSHUN" (mixed).
    • "Where you left off" (Continuity rail).
    • Domain-specific lane (driven by filter).
    • "Editor's notes" (manually curated).
    • "Slow recommendations" (longer-form, 2+ minute reads).
  4. Lane card. Image anchor 16:9, title (title-md), source/author chip, confidence/evidence chip (Veritas), domain accent dot.

Empty state. "Quiet today. Try a domain filter or ask the assistant." with subtle prompt.

Anti-goals. No infinite-scroll algorithmic feed. No autoplay video.


E.6 Customer shell — Activity#

Goal. Surface the user's recent meaningful progress and gentle next steps. Not a productivity audit.

Layout:

  1. Streak summary header — small stat tile per domain (only for active streaks). No rank/leaderboard. Streak design avoids "you broke your streak" shame.
  2. Timeline — chronologically reversed list of recent cross-domain actions. Each row: timestamp (label-sm), domain accent dot, action description (body-md), optional inline action ("Resume," "Re-read," "Recover"). Group by day.
  3. Recover/retry section. Failed background actions (saved-but-not-synced, session interrupted, etc.) appear with retry CTAs.

Recovery copy. Always supportive ("Pick this up when you can"), never punitive. After a missed day, surface explicit recovery prompt: "You missed yesterday. Want to do today's check-in now?"


E.7 Customer shell — Library#

Goal. Unified saved/bookmarked items across all domains, organised so the user can find them by domain, by collection, or by recency.

Layout:

  1. View tabs: "All · Tara · Arete · Veritas · Nyx · Nisaba · Collections."
  2. Sort/filter row: date saved · domain · type · "with notes."
  3. Item grid/list toggle. Default: vertical list with media thumb + metadata.
  4. Collections panel (right side on desktop, bottom sheet on mobile). Custom user-named collections. Drag/drop on web; long-press menu on mobile.

Item card. Thumb (1:1 or 16:9), title (title-md), source/domain dot, saved date (label-sm), inline actions: open, remove, move to collection.

Empty state. Per domain, a single editorial line: "Nothing saved from Nyx yet. Tonight's sky is a good place to start."


Goal. One search across the five domains with continuity, evidence, and preview.

Layout (desktop overlay, opens with cmd-K):

text
┌──────────────────────────────────────────────────────────────────────┐
│  [search input — large, focus-ringed]                          ✕     │
├──────────────────────────────────────────────────────────────────────┤
│  Continuity strip (3 most-recent items, horizontal)                  │
├──────────────────────────────────────┬───────────────────────────────┤
│  Results column (denser list)        │  Preview / evidence lane      │
│  · Result 1 — domain dot + title     │  Selected result preview      │
│  · Result 2                          │  Source lineage (Veritas)     │
│  · Result 3                          │  Confidence chip              │
│  ...                                 │  Inline actions (open, save)  │
└──────────────────────────────────────┴───────────────────────────────┘

Mobile. Full-screen, no two-pane. Selecting a result pushes preview view.

Domain filter chips above results. Domain accent dots in result list.

No-results state. "Nothing matches. Try the assistant?" → opens Assistant.


E.9 Customer shell — Profile / Settings#

Goal. Calm, complete control surface. User can find and change anything about account, privacy, accessibility, theme, notifications, and connected domains in under three taps.

Sections (vertical, scannable):

  1. Account. Name, email, password, sign-in methods, sessions, devices.
  2. Subscription / billing. Tier, next charge, manage.
  3. Privacy & data. Memory disclosure, data export, deletion, consent audit, telemetry preferences.
  4. Accessibility. Reduced motion, high contrast, large text, screen reader optimisations, focus ring boost, audio descriptions, captions on ambient media.
  5. Theme. Default (dark) / Light / Auto. High-contrast variant.
  6. Notifications. Per-domain delivery rules, quiet hours, push channel selection.
  7. Connected domains. Toggle which rooms appear; reorder.
  8. About / legal. Version, terms, privacy policy, support.

Row pattern. Label (body-md) · current state (label-md, fog-300) · chevron. Toggles on right, ≥48dp.


E.10 Tara — Meditation Room#

Goal. Get the user into a session in ≤2 taps from the Tara entry. Settle the body before the screen does.

Surfaces in Tara:

  1. Tara Home. Continuity card (resume current course / return to today's ritual), session quick-start strip (3-min, 7-min, 15-min), library lanes (calm, focus, sleep, breath), one editorial feature.
  2. Session player. Full-screen. Slow ambient still or video at low amplitude. Dim lower band with: title, instructor name, transport controls (play/pause, skip 15s, scrub bar with time labels), duration, single inline AI-disclosure label if AI-narrated, exit affordance.
  3. Pre-session screen. "Are you ready?" — 8s grounding moment. Optional skip. Tap-to-continue.
  4. Post-session screen. "How did that feel?" — three quiet emoji-free chips (calmer, same, restless). Save → Activity. No streak shame on "restless."
  5. Course detail. Cover image, course meta, sessions list with state (started/done/locked).
  6. Instructor profile. Photo (real, editorial), bio, sessions taught, teaching philosophy single paragraph.
  7. Breath ritual. A focused breathing UI — animated circle expanding inhale/exhale on a 4-7-8 or box-breath cadence. Reduced-motion replaces animation with text cue.

Domain accent. domain-tara #2BC6B4 on the player progress bar, on "Tara" labels, and on the pre-session pulse ring.

Anti-goals. No karma points. No "session #57." No emoji "How do you feel today?" pickers.


E.11 Arete — Goals, Habits, Journal, Coach#

Goal. Convert intention into ≤60s daily check-in and visible progress without guilt.

Surfaces:

  1. Arete Home. Daily check-in card (top, primary), habit row (3 habits max as chips with state), today's journal prompt teaser, coach summary card, weekly review entry.
  2. Daily check-in screen. Three vertical cards: "How did yesterday go?" (3 quick chips), "What's one thing for today?" (single-line input), "Anything to log?" (optional). One scroll. Submit ≤60s for power user.
  3. Habits view. Each habit row: name, current streak (compact), today's state toggle, trend chart (sparkline, last 7 days). Edit on long-press / right-click.
  4. Journal. Two modes: prompted (today's prompt + composer) and freeform (blank composer). Markdown supported. Auto-save every 3s. AI coach can summarise on request (optional, never automatic).
  5. AI coach panel. Conversational surface. Coach tone is calm, never exhortative. Answers ground in user's recent journal/habit data. Memory disclosure visible.
  6. Weekly review. Slow scroll layout: streak summary, journal highlights, habit chart, "what shifted?" reflection prompt, set-next-week intentions.

Domain accent. domain-arete #3AC77B on streak rings, today-completed states, primary CTAs within Arete views (overrides aqua only inside Arete chrome).

Recovery. After missed day, the daily check-in screen leads with: "It's okay. Pick today up." No counter resets visible until user explicitly views streak settings.

Anti-goals. No leaderboards. No XP. No "you've earned the diligence badge." No animated trophy.


E.12 Veritas — Verified Reading#

Goal. Make the user feel confidently informed without manipulation. Evidence is visible without making the surface feel clinical.

Surfaces:

  1. Veritas Home. Today's verified stories rail, in-progress claim verifications, saved-for-later list, reading queue.
  2. Story / claim reader. Long-form reading frame (62-char measure, body-lg 17–18px). Right-side persistent evidence rail showing: claim confidence chip (high / medium / contested), source list (with credibility signals), inline citation jump-to. On mobile, evidence rail becomes a bottom drawer.
  3. Source lineage view. Document tree of sources, primary sources flagged, credibility weighting, date and provenance. Quiet, factual treatment.
  4. Claim checker. Input field "Paste a claim, sentence, or URL." Returns confidence, supporting/contradicting evidence with source rank, history of the claim.
  5. Topic hub. Collected stories around an unfolding topic; timeline view.
  6. Reading queue. Saved items in a serial reader with progress.

Confidence chip design.

text
[High confidence ●]   green-ish, success-500 dot, label-md
[Medium confidence ●] amber-500 dot
[Contested ●]         danger-500 dot, paired with explanation toggle
[Unverified ●]        fog-300 dot

Domain accent. domain-veritas #27A3F3 on evidence rail edge, citation jump-to dots, primary CTA within Veritas.

Anti-goals. No "BREAKING" red chyron. No fear-based icons. No source shame. No partisan visual coding.


E.13 Nyx — Sky, Calendar, Observation#

Goal. Turn curiosity into a regular evening ritual. Wonder + competence, even outdoors with a phone in low light.

Surfaces:

  1. Nyx Home. "Tonight's sky" hero (single dominant ambient field — slow star canvas), top events strip, observation windows for current location, recently viewed objects.
  2. Sky map. Interactive star map. Pinch zoom, drag, tap object → object detail. Constellation lines toggle. Real-time altitude/azimuth based on user location. Compass and time-slider controls.
  3. Object detail. Object name, image, current rise/set, notable characteristics, short awe-and-perspective context (one paragraph), related Nisaba reading, save.
  4. Celestial calendar. Month view; events plotted; tap event → detail with observation window.
  5. AR sky (browser AR or native). Camera passthrough with overlaid constellation lines and labels. Calibration step on first use.
  6. VR planetarium (if WebXR enabled).

Night mode (mandatory). Auto-engages after sunset based on local time. Switches to deep red / amber chromatics for legibility without ruining dark adaptation. UI brightness reduces; accent colours desaturate.

Domain accent. domain-nyx #7D7CFF on event chips, sky-map UI controls, object detail accents. In night mode, accent shifts to a warm amber-red (adapted token) to protect dark adaptation.

Anti-goals. No HUD-style overlays. No neon. No starship gradients. Avoid generic space wallpapers.


E.14 Nisaba — Long-form reading desk#

Goal. Editorial, durable reading surface. The most text-forward domain. Calm enough to read for an hour.

Surfaces:

  1. Nisaba Home. Continue reading rail, daily passage card, study plan progress, lexicon search entry, notebook recent entries.
  2. Reading desk. The core surface. Three-pane on desktop:
    • Centre (62-char): passage body. Drop-cap optional. 1.6 line-height. body-lg 18/28.
    • Left (compare slip, optional): a second passage variant (translation, edition).
    • Right (notes / lexicon): annotation thread, lexicon hover panel, concept jump-to.
    • On mobile, panes collapse to drawers; reader is full-bleed.
  3. Lexicon panel. Tap any word → morphology, etymology, frequency, related passages.
  4. Concept graph. Interactive node-link explorer of related concepts.
  5. Notebook. Notes organised per passage, per concept, per study plan. Supports inline quotes, highlights, custom tags.
  6. Study plan view. Reading list with progress bars, scheduled passages, learning objectives.

Reading frame standards.

  • Measure: 62ch.
  • Line-height: 1.6.
  • Paragraph spacing: 24px.
  • Justification: left-aligned, never full-justified.
  • Pull quotes: serif (Space Grotesk allowed for emphasis), display-sm size, 10% page-margin indentation.

Domain accent. domain-nisaba #C88B2C on highlight pin, lexicon panel edge, scholarly chrome.

Anti-goals. No skeumorphic book illustrations. No fantasy parchment textures. No "reading achievement unlocked."


E.15 Notifications centre#

Goal. Quiet inbox of system + domain notifications with strong user control.

Layout:

  1. Tabs: All · Unread · By domain.
  2. Row pattern: domain dot · title (body-md) · preview (body-sm, text/secondary) · timestamp (label-sm). Inline actions: dismiss, snooze, mute domain.
  3. Empty state: quiet checkmark + "All caught up."

Notification preferences sub-screen. Per-domain delivery rules: push/inapp/email checkboxes, priority threshold (all / important / silent), quiet hours.

Anti-goals. No badge bombardment. No red-dot anxiety designs.


E.16 Assistant Panel (global overlay)#

Goal. Always-available AI assistant for cross-domain questions, continuity journeys, and domain handoffs. Memory and disclosure are first-class.

Layout:

  1. Header. "OSHUN Assistant" + memory disclosure toggle ("Using your recent activity") + close.
  2. Conversation thread. User and assistant messages. Assistant messages show domain handoff chips when surfacing domain content ("From Veritas: …"). Confidence chip when relevant.
  3. Input bar. Single-line composer + voice input + attach context (recent activity, an item, a note).
  4. Continuity rail (top of input). Two or three suggested contextual prompts ("Resume Tara session," "Today's claim," "Read passage 4 of 5").
  5. AI disclosure. Persistent footer note: "Assistant uses recent activity to ground replies. Open settings to control memory."

Domain handoff chip pattern. When the assistant surfaces domain content, the message bubble has a thin domain accent left edge and a chip "From [Domain]" that links to that domain's view.

Anti-goals. No anthropomorphic avatar (no big animated face). No exclamation-mark conversational style. No hidden persistent listening indicator without explicit user action.


E.17 Domain Switcher (mobile bottom sheet)#

Goal. Switch between rooms in ≤1 thumb-reachable tap from Home, without losing context.

Layout (bottom sheet):

  1. Five domain rows, top-to-bottom, each with: domain accent left edge, icon, name, current continuity teaser ("Continue: 7-min Calm Drift"), inline CTA (Open / Resume).
  2. Below: "Switch with continuity" toggle (default on).
  3. Drag handle and dismiss.

Trigger. Bottom-left FAB on Home tab; cmd-D on web (desktop has sidebar launcher built in).


E.18 Empty / loading / error / offline / partial-failure states#

These are mandatory across all surfaces. Generate a state library Claude Design can instantiate per surface.

text
Loading        Skeleton with restrained shimmer (linear, 1200ms loop, 8% alpha
               highlight band). No spinning circles for primary content.

Empty          One short editorial line + one optional CTA. No mascots.
               Example: "Nothing here yet. Today's sky is a good place to
               start."

Error          Quiet card with icon (16px, fog-300), one-sentence cause,
               retry CTA, secondary "Tell us about this" link. Never red
               background. Never modal blocking unless destructive.

Offline        Top inline banner ("Offline. We'll resync when you're back.")
               + offline-cached content marked with a small dot.

Partial        Per-domain card shows quiet "Tara unavailable. Try again
failure        shortly." inside the card frame; rest of shell remains live.

Reduced        All ambient motion freezes, route transitions ≤100ms, list
motion         staggers off, breath animations replaced by text cue.

E.19 Email templates#

Generate a single email template system covering: account verification, password reset, weekly digest, domain handoff prompt, billing receipt.

Style. Editorial, calm. Plain typography (Manrope on web-safe fallback), ink-950 on white, single column 600px max width. One image anchor at top optional. Footer: small wordmark + unsubscribe + legal.

No. No big coloured CTA buttons that look like SaaS marketing. No "Hey [Name]!!" ALL-CAPS subject. No multi-column layouts.


E.20 Open Graph + share cards#

For each domain, generate a social-share card template. 1200×630.

  • Tara cards: dark navy gradient, soft moonlit asset, single line of copy in Space Grotesk, domain accent under title.
  • Veritas cards: documentary photograph or single source quote with citation, confidence chip overlay.
  • Nyx cards: sky photograph with object name overlay, observation window text.
  • Arete cards: typographic, no image, single intention sentence in display-lg.
  • Nisaba cards: manuscript-margin texture with a passage opening line in serif italic.

E.21 Admin web — Operator shell#

Goal. Calm, dense operator surface for review queues, incidents, content governance, and policy control. Different visual posture from customer shell — denser, more data-tabular — but same brand voice.

Customer/admin distinction. Admin surfaces may show greater density and more simultaneous controls than customer. They share tokens but use title-lg instead of display-lg, never the customer canvas gradient, and the chrome is ink-900 instead of nocturnal gradient.

Admin vs. Studio vs. Tenant Console. Three distinct operator products share the operator-grade visual posture. Admin web (this section) is a focused operator triage cockpit — Inbox, Queues, Incidents — for fast review of items the platform produced. Studio (E.23–E.24, E.26) is the much-larger authoring + governance + pipeline product (hundreds of workspaces). Tenant Console (E.25) is the tenant-scoped admin product that institutions use to manage their members, identity, content scope, and compliance. All three obey the same density rules; Admin is the densest and most triage-focused.

Top-level workspaces (left rail):

  1. Inbox — cross-workspace triage.
  2. Queues — Bellona render queue, content review, safety review.
  3. Incidents — severity-bucketed, SLA timers, acknowledge/coordinate/close.
  4. Governance — quick-action shortcuts to Studio governance workspaces (token publish, design language, taxonomy, lexicon). Studio itself is E.23.
  5. Tenants — quick-action shortcuts to Tenant Console (E.25) per tenant; plus cross-tenant audit views.
  6. Analytics — KPI dashboards (V1 customer KPI dashboards exist; reuse).
  7. Settings — admin user account, notifications, audit log.

Queue row pattern.

text
[priority dot] [type icon] Title (body-md)         ⋅ submitter ⋅ age ⋅ SLA timer pill ⋅ assignee ⋅ [actions]

Incident row pattern.

text
[severity badge S1/S2/S3] Title ⋅ status (Open/Acked/Mitigating/Closed) ⋅ ETA ⋅ owner ⋅ link

SLA timer pill. Time-remaining in label-sm, colour scales: green ≥30%, amber 10-30%, red <10%. Pulse animation only on red and only at 200ms intervals for accessibility.

Policy diff panel. Side-by-side diff (similar to GitHub PR diff but calmer, with inline annotations).

Anti-goals. No customer-style "calm hero." No domain accents on shell chrome. No nocturnal gradient. Density is fine, anxious density is not.


E.22 Admin mobile — Urgent ops companion#

Goal. On-call companion for incident acknowledgement and urgent queue intervention. Push-driven.

Tabs (bottom): Urgent · Inbox · Incidents · More.

Urgent tab. Single-screen list of breaching SLAs and escalations; acknowledge in one tap. Step-up auth gate before destructive actions.

Incident detail. Severity badge prominent, status, owner, timeline of events, action buttons (acknowledge, coordinate, close).

Step-up auth. Biometric or PIN before high-risk actions; clear explanation ("This decision will be audited. Confirm with biometric.").


E.23 Oshun Studio — Shell + Universal Workspace Template#

Goal. Studio is the operator/creator/governance product for Oshun. It is a single web app that hosts hundreds of governance, pipeline, and authoring workspaces under one shell. The visual posture is distinct from the customer shell and the admin shell: operator-grade, dense, low-noise, keyboard-first — "an air-traffic control room for governed intelligence work."

Visual thesis (canonical).

Operator-grade, low-noise cockpits with dense evidence, durable context, and keyboard-first ergonomics.

Reference posture. Linear (keyboard-first nav, fast list scanning, quiet sidebar command structure) and Stripe Dashboard / Workbench (embedded diagnostics, evidence-rich logs, operator tooling that stays inside one cockpit). Posture only — do not imitate brand or layout cosmetics.

How Studio differs from customer shell:

text
                    Customer shell        Studio shell
Canvas              Nocturnal gradient    Neutral ink-900 flat
First impression    Calm hero              Dense list / queue
Type ceiling        display-xl             title-xl
Density             Spacious               Tight rows (≤44px)
Domain accents      Atmosphere in rooms    Severity / state only
Motion              Settling, breath       Confirm + handoff (≤200ms)
Keyboard            Optional               Mandatory shortcuts
Card mosaics        Off-thesis             Off-thesis (rows beat tiles)

How Studio differs from admin shell (E.21):

Admin shell handles review/incident triage (operator-side decisions on content already produced). Studio handles authoring + pipeline + governance configuration (the controls that produce or shape the content). Both share operator visual posture, density rules, and keyboard ergonomics; Studio adds multi-pane workspace canvases and live pipeline state.

E.23.1 Studio shell layout#

text
┌──────────────────────────────────────────────────────────────────────────────┐
│ Top bar (40px): org / tenant switcher · workspace search · cmd-K · user      │
├────────────────┬─────────────────────────────────────────────────────────────┤
│                │                                                              │
│  Workspace     │  Workspace canvas                                           │
│  Navigator     │                                                              │
│  (260px)       │  Each workspace renders inside this canvas using the        │
│                │  universal workspace template (E.23.2). Sticky workspace    │
│  Foundation    │  header pins context. Body splits into nav / detail /       │
│  Identity      │  inspector panes per workspace type.                         │
│  Collaboration │                                                              │
│  Operations    │                                                              │
│  Reliability   │                                                              │
│  Pipelines ›   │  ┌── Workspace header (sticky, 56px) ─────────────────────┐ │
│   Bellona      │  │ [Workspace name]  ⋅  [context chip]  ⋅  [tenant pill] │ │
│   Hathor       │  │ [actions: publish · review · simulate · …]            │ │
│   Yemaya       │  └────────────────────────────────────────────────────────┘ │
│   Aja          │                                                              │
│   Isis         │  Tabs: Overview · Journeys · Evidence · Audit · Settings   │
│   Neith        │                                                              │
│  Authoring     │  [Workspace canvas — see E.23.2]                            │
│   Concordia    │                                                              │
│   Obsidian     │                                                              │
│  ────          │                                                              │
│  Recent        │                                                              │
│  Pinned        │                                                              │
│                │                                                              │
└────────────────┴─────────────────────────────────────────────────────────────┘

Workspace navigator groups (canonical IA):

text
Foundation         Information Architecture, Design Language, Typography,
                   Color System, Spacing/Layout, Component Primitives,
                   Complex Interactions, Navigation/Commands,
                   Search/Discovery
Identity           Authentication Architecture, Session/Device Management,
                   RBAC/Permission Policy, Enterprise Tenant Isolation,
                   Audit/Compliance Surfaces
Collaboration      Real-Time Collaboration Substrate, Presence/Cursor
                   Systems, Commenting/Annotation System,
                   Review/Approval Workflows, Activity/Change Feeds,
                   Notification Center
Operations         Cross-Domain Entity Model, Workspace Context Switching,
                   Multi-Project Operations, File/Media Ingestion,
                   Asset Preview Pipeline, Background Jobs/Progress UX,
                   Resilience/Error UX, Webhooks/External Automation,
                   API Gateway/BFF Composition, SDK/Documentation
Reliability        Metrics/Analytics Instrumentation, Observability/
                   Operational Dashboards, Performance Budgets,
                   Experimentation/Feature Flags, Accessibility Governance,
                   Internationalization/Localization, Security Hardening,
                   Data Retention/Lifecycle, Backup/DR UX,
                   Launch Readiness Governance
Pipelines          Bellona, Hathor, Yemaya, Aja, Isis, Neith
                   (each expandable to its own sub-tree of pipeline
                    workspaces — see E.24)
Authoring          Concordia Workbench, Project Obsidian

cmd-K command palette. Universal across Studio. Searches workspaces, recent items, jumps to actions ("publish design language v2," "open Bellona render queue," "simulate RBAC decision"). Keyboard-first.

E.23.2 Universal workspace template#

Every Studio workspace conforms to one of two layout templates. Pick by workspace type:

Template A — Governance / policy / operations workspace (3-pane, used by ~80% of workspaces):

text
┌──────────────────────────────────────────────────────────────────────────┐
│ Workspace header (56px sticky):                                          │
│   [Workspace name] · [context chip: tenant/project/release] · [actions]  │
├──────────────────────────────────────────────────────────────────────────┤
│ Tabs: Overview · Journeys · Evidence · Audit · Settings                  │
├────────────┬───────────────────────────────────┬─────────────────────────┤
│ Navigator  │  Detail canvas                    │  Inspector              │
│ (240px)    │  (fluid)                          │  (320px)                │
│            │                                    │                         │
│ List of    │  Selected item view: structured   │  Context-bound details  │
│ items      │  fields, evidence panels, action  │  for selected item:     │
│ (queue,    │  controls, validation states.     │  policy, ownership,     │
│ policies,  │                                    │  audit links, related   │
│ rules,     │  Sub-tabs inside detail when     │  decisions, lineage,    │
│ entities)  │  needed (e.g., Inputs · Outputs   │  evidence pointers.     │
│            │  · History).                       │                         │
│ Filters    │                                    │  Persists across item  │
│ + density  │                                    │  selection so operator │
│ controls   │                                    │  doesn't lose context. │
│            │                                    │                         │
│ ────       │                                    │                         │
│ Bulk       │                                    │                         │
│ actions    │                                    │                         │
└────────────┴───────────────────────────────────┴─────────────────────────┘

Template B — Authoring / pipeline canvas workspace (used by Bellona/Yemaya/ Hathor/Aja/Isis pipeline workspaces, Concordia, Obsidian):

text
┌──────────────────────────────────────────────────────────────────────────┐
│ Workspace header (sticky)                                                │
├──────────────────────────────────────────────────────────────────────────┤
│ Toolbar: select · pan · zoom · grid · simulate · export · run            │
├────────────┬──────────────────────────────────────────────┬──────────────┤
│ Asset/     │  Authoring canvas (graph, timeline, 3D       │  Properties  │
│ Library/   │  preview, render preview, etc.)              │  (320px)     │
│ Tree       │                                               │              │
│ (260px)    │  Examples by workspace family:                │  Selected    │
│            │   - Bellona render queue: timeline of jobs    │  item attrs, │
│ Hierarchy  │   - Hathor world editor: map + entity tree    │  policies,   │
│ navigation │   - Yemaya project: asset graph + timeline    │  validation  │
│            │   - Aja motion pipeline: clip + skeleton view │  panel,      │
│            │   - Isis workflow: ComfyUI-style node graph   │  AI assist   │
│            │                                               │              │
│            │  Bottom dock (collapsible): logs · timeline · │              │
│            │  diagnostics · live preview                   │              │
└────────────┴──────────────────────────────────────────────┴──────────────┘

E.23.3 Per-workspace canonical structure#

Every workspace, regardless of template, exposes these five tabs:

text
Overview     Top-level state: queue health, release posture,
             tenant context, current decision posture, "what needs
             my attention now."

Journeys     The named operator journeys this workspace governs
             (e.g., for Tenant Isolation: J1 Tenant Context Selection,
             J2 Tenant-Scoped Policy Decision, J3 Cross-Tenant Denial,
             J4 Boundary Violation Recovery, J5 Policy Promotion/Rollback).
             Each journey shows current step, progress, and decision
             checkpoints.

Evidence     Structured evidence panels: validation runs, audit
             trails, decision logs, policy diffs, source data
             pointers. Each evidence row has timestamp, actor,
             outcome, link.

Audit        Append-only timeline of every decision/state change in
             this workspace. Filterable by actor, action type, outcome.
             Export as compliance evidence bundle.

Settings     Workspace-scoped configuration: SLAs, escalation rules,
             notification routing, feature flags, retention policy.

E.23.4 Studio chrome conventions#

  • Density. Rows 36–44px, padding 8–12px. Operator can opt into "comfortable" (44–52px) via density toggle.
  • Tables. Sticky header, sortable columns, column resize, frozen first column for wide tables, virtualised scroll for >100 rows.
  • Severity badges. S1 danger-500, S2 amber-500, S3 fog-300, label-sm, weight 600.
  • Status pills. Open · In Progress · Pending Review · Approved · Blocked · Failed · Closed. Status-specific dot colour; never a full-pill colour wash.
  • Decision chips. Allow success-500 outline · Deny danger-500 outline · Defer fog-300 outline. Outlines, not fills, to avoid alarm chrome.
  • Diff panes. Side-by-side diff with inline change markers. Removed lines in danger-500/12% background, added in success-500/12%. Word-level diff highlighting in the inline mode.
  • Validation states. Inline pill — "Validating…" (spinner), "Validated ✓" (success-500), "Validation failed" (danger-500 + click for details).
  • Keyboard shortcuts. ? shows shortcut help. Common shortcuts:
    • j/k — next/prev item in navigator
    • enter — open selected item
    • cmd-k — command palette
    • cmd-/ — workspace search
    • g then o/j/e/a/s — jump to Overview/Journeys/Evidence/Audit/Settings
    • cmd-shift-d — toggle density
    • cmd-shift-i — toggle inspector
    • e — edit · r — review · p — publish · x — escalate

Anti-goals.

  • Customer-style hero banners on operator surfaces.
  • Dashboard-card sprawl where rows or panes scan faster.
  • Permanent red/orange atmosphere (alert fatigue).
  • Hidden critical context behind multiple clicks.
  • Modal-heavy workflows that fragment review context.
  • Glossy motion or parallax in routine workflows.
  • Domain accents on Studio chrome (severity colours only).
  • Decorative entrances or theatrical transitions.

E.24 Studio — Domain pipeline workspace families#

Studio hosts five domain pipeline families, each with dozens of subworkspaces. They share Template B (authoring canvas) and the universal Overview/Journeys/ Evidence/Audit/Settings tabs, but each family has distinct content and one or two family-specific surface patterns.

E.24.1 Bellona — Engine bridges, build, render, virtual production#

What it does. Real-time engine integrations (Unity/Unreal/Godot/Blender/ Houdini/DaVinci), USD pipeline, MetaHuman, XR (visionOS, Meta Quest), motion capture streaming, virtual production (LED wall, ICVFX, genlock), build and render queue management.

Family surfaces:

  • Live Engine Bridges hub. Connection state per engine target, session count, throughput, last error. Tap a bridge → bridge detail (commands, protocol, sessions).
  • Bridge command surface. Send/receive command list with timestamps, payload preview, replay button.
  • Render queue. Job list with [priority dot] [type] Title · submitter · age · ETA · status · render farm node · GPU class. Filters: status, type, submitter, project. Per-job detail: job graph, asset inputs, output manifest, log tail, retry/cancel actions.
  • Build queue tracking + artifact management. Build pipeline graph (nodes: fetch → compile → package → sign → publish), artifact list with content hash, size, target, retention.
  • OpenUSD stage. 3D viewport with stage hierarchy (left), USD layer stack (right), variant manager (bottom dock).
  • MetaHuman pipeline. Avatar identity preview, rig binding inspector, retargeting parameters.
  • XR platform support. Per-platform feature matrix (visionOS / Meta Quest / Shared XR), capability test results, build target settings.
  • Virtual production. LED wall calibration, ICVFX scene assembly, genlock / timecode monitor.

Family-specific surface patterns:

  • 3D viewport. WebGL/WebGPU canvas; camera controls (orbit/pan/zoom); hierarchy tree on left; properties on right. Reduced-motion disables auto- orbit; static frame fallback.
  • Render preview strip. Horizontal scroll of frame thumbnails with scrubber; click frame → full-resolution preview with diagnostics overlay (GPU time, memory, draw calls).

E.24.2 Hathor — Worldbuilding, narrative, simulation#

What it does. World/character/faction/location modeling, NPC behavior trees, dialogue, NPC memory + emotion + safety guardrails, narrative graphs, simulation (economy/politics/culture), lore validation, screenplay compilation.

Family surfaces:

  • Workbench dashboard. Active worlds, recent edits, validation health, cross-domain integration status (Sophia citation, Bellona compilation).
  • World editor. Map canvas (left), entity hierarchy (left rail), entity properties (right). Timeline scrubber at bottom for temporal world state.
  • Character modeling. Bio · personality · emotional state · memory · relationships tabs; consistency validation panel (contradictions detected).
  • Story graph authoring. Node graph: scenes/beats/branches; click node → prose body, entity refs, lore citations. Lint panel highlights breaks in causality, taxonomy, or contradiction.
  • Dialogue tree. Tree canvas; node properties (speaker, intent, preconditions, postconditions); inline NPC personality preview.
  • Simulation runners. Economy/politics/culture simulation: parameters panel, run button, time-series result chart, snapshot diff vs prior run.
  • Unified validation dashboard. All lore validators (timeline, causality, taxonomy, contradictions) on one screen with per-validator pass/fail count and drill-down.

Family-specific surface patterns:

  • Lore validation pane. A persistent right-rail mode showing currently active validators and live error count as you author.
  • Cinematography planning timeline. Horizontal beat strip with shot/coverage thumbnails.

E.24.3 Yemaya — Creative production studio#

What it does. Multi-asset creative project management for film/game/XR production: assets, projects, team collaboration, screenplay/storyboard, character/world design, integrations (Blender/Godot/Unreal/Houdini/DaVinci), multi-agent orchestration, autonomous film/game pipelines, marketplaces, enterprise org/RBAC/compliance, real-time presence, voice/video, scheduling/ budget/crew/location management.

Family surfaces:

  • Project home. Overview of project state: phase, deadline, asset count, team, recent activity, autonomous-pipeline state.
  • Asset library. Grid + list dual mode; filter by type/status/owner. Selection → asset detail (preview, metadata, lineage, dependents).
  • Asset pipeline canvas. Node graph of asset transformations.
  • Team collaboration. Real-time presence cursors on shared canvas; voice/ video pip; comments thread; CRDT sync indicator.
  • Storyboard creation. Beat-cell grid with image, shot description, duration, dialogue snippet.
  • Autonomous pipeline. Pipeline templates list; run history; per-run graph view of agent actions, decision points, human-in-loop checkpoints.
  • Marketplace. Plugin / AI model / asset listings; per-listing detail with reviews, license, version history.
  • Enterprise admin. Org members, RBAC, compliance posture, analytics dashboard.

Family-specific surface patterns:

  • Real-time presence cursor layer. Coloured cursor labels visible on any shared canvas; click cursor → user profile.
  • Voice/video PiP. Floating draggable PiP frame with current speakers, mute/camera controls.

E.24.4 Aja — Motion capture and embodied instruction#

What it does. Motion-capture pipeline: pose estimation, retargeting, retiming, gap filling, quality scoring, domain-specific pipelines (yoga, fitness, dance, martial arts, sports, medical/rehab, film/VFX, gaming), avatar library, IK, batch operations.

Family surfaces:

  • Pipeline canvas. Node graph: input video → pose extract → lift to 3D → retarget → retime → quality check → output. Each node has params, run state, diagnostic.
  • Skeleton viewer. Animated 3D skeleton viewport with playback controls; side-by-side ground truth comparison.
  • Avatar library. Avatar grid with skeleton type, rig quality score, preview.
  • Domain pipeline. E.g., Yoga pipeline — pose-detection accuracy, named asana classifier, alignment notes for instructional capture.
  • Quality dashboards. Basic + perceptual metrics chart; ground truth diff; automated QA pipeline result.

Family-specific surface patterns:

  • Skeleton viewport with overlay. Source vs retargeted skeleton in different colours; toggle bones, joints, IK targets.
  • Frame quality timeline. Strip of frame thumbnails coloured by quality score; click → frame inspector.

E.24.5 Isis — Generative AI workflows#

What it does. Image / video / 3D / audio generative pipelines: ComfyUI integration, multi-provider routing, workflow versioning, model registry (Civitai, LoRA training, model merging), provenance tracking, content safety, content watermarking, retention policies, budget/cost management.

Family surfaces:

  • Workflow canvas. ComfyUI-style node graph (nodes draggable; sockets type-checked; live preview when running). Node palette on left.
  • Job submission. Form-driven for simple cases; canvas-driven for workflow authoring. Submitted jobs land in job-management queue.
  • Job management. Queue list with status, ETA, GPU class, cost; click → job detail (inputs, outputs, logs, manifest, lineage).
  • Output gallery. Grid of generated assets with metadata, provenance link, watermark indicator.
  • Provider registry. Provider list with capability matrix, cost, latency, reliability; routing rule editor.
  • Model registry. Model list with version, size, source (Civitai/internal), quality score, usage stats.
  • Provenance tracking. Lineage graph for any output: model → workflow → job → output, with policy posture and watermark status.
  • Cost dashboard. Spend by provider/project/user, budget remaining, optimization recommendations.
  • Content safety. Anomaly detection results, suspicious activity log, account protection state.

Family-specific surface patterns:

  • Node graph editor. Pannable/zoomable canvas; minimap; node search; selection multi-select with bulk param edit.
  • Live preview pane. Output renders incrementally as the workflow runs; step-by-step intermediate previews available.

E.24.6 Common pipeline-canvas patterns (apply to all five families)#

  • Run button sticky to top of canvas; state: idle / queued / running / paused / failed / done. Click to run; cmd-enter shortcut.
  • Live diagnostic dock at bottom (collapsible): logs, timeline, GPU/CPU metrics, error tail.
  • AI-assist drawer on right (toggleable): conversational helper that can suggest pipeline improvements, fix errors, generate node configurations.
  • Validation panel persistent: structural validation (required inputs bound), policy validation (allowed providers/budgets), output validation (quality threshold met).

E.25 Tenant Console — Institution / Tenant Administration#

Goal. Tenant administrators of institutions, schools, research labs, partner orgs need their own scoped admin surface to manage members, identity integrations, content scoping, audit, compliance, billing, and lifecycle. This is distinct from Studio (which is the Oshun-org governance product) and distinct from Admin (which is the Oshun internal operator product).

Tenant Console personas:

  • Tenant Administrator — owns tenant configuration, members, billing.
  • Identity Operator — runs SSO/SCIM, provisions users, audits sign-ins.
  • Compliance Auditor — exports audit, runs compliance reviews.
  • Content Curator — scopes domains, configures content filters, manages cohorts.
  • Billing Owner — plans, invoices, usage, seat allocation.

Tenant Console differs from Studio:

text
                    Studio                    Tenant Console
Audience            Oshun org operators        Tenant (institution) admins
Scope               All tenants, governance    One tenant, configuration
                    of platform behaviour      of tenant behaviour
Density             Operator-grade dense       Slightly less dense
                                               (fewer concurrent decisions)
Visual              Air-traffic control        Calm administrative panel
                    cockpit                    (between Studio density and
                                               customer calm)
Branding            Oshun chrome only          Tenant-branded shell allowed
                                               (logo, accent colour) within
                                               Oshun visual grammar

Top-level tenant console workspaces:

text
Overview         Tenant health dashboard: members, active seats, recent
                 invitations, sign-in volume, recent changes, alerts.

Members          Roster: name, email, role, status (active/invited/
                 suspended), last sign-in, groups. Bulk import via CSV /
                 OneRoster. Per-member detail: role assignments, group
                 membership, sign-in history, devices, audit trail.

Identity         SSO (SAML, OIDC) configuration; SCIM provisioning;
                 OneRoster (for education tenants); domain whitelisting
                 for self-serve sign-up; MFA enforcement policy.

Roles &          Tenant-scoped role catalogue (Tenant Admin, Curator,
permissions      Member, Auditor, etc.); permission matrix; role
                 simulation ("if this user does X, what happens?").

Content scope    Which Oshun domains / Studio workspaces this tenant has
                 access to. Per-domain content filters. Cohort definitions
                 (e.g., "Grade 9 students," "Research cohort A").

Cohorts &        Group-level configuration: enrolment, content
groups           assignments, schedule, pacing.

Audit            Tenant-scoped audit log: every member action, identity
                 event, configuration change. Filter by actor, action,
                 date. Export as compliance bundle (PDF, JSON, CSV).

Compliance       Posture dashboard: GDPR/CCPA/COPPA/FERPA status (varies
                 by tenant type), data retention policy, data export
                 requests, deletion requests, consent records.

Integrations     Connected third-party apps; webhooks; API keys with
                 scope and rotation history.

Billing          Plan, seats, usage, invoices, payment method, alerts on
                 usage thresholds.

Branding         Tenant logo, accent colour (within Oshun palette), email
                 sender display name. Limited branding — Oshun chrome
                 stays primary.

Settings         Tenant name, primary contact, default language, default
                 timezone, default theme, support contact.

E.25.1 Tenant Console layout#

Same shell pattern as Studio but with the tenant brand visible:

text
┌──────────────────────────────────────────────────────────────────────────┐
│ Top bar: [Tenant logo] Tenant Name · workspace search · cmd-K · admin    │
├────────────┬─────────────────────────────────────────────────────────────┤
│            │                                                              │
│ Workspaces │  Workspace canvas                                           │
│ (240px)    │                                                              │
│            │  Most workspaces use Template A (3-pane: navigator / detail │
│ Overview   │  / inspector). Members and Audit are Template A. Compliance │
│ Members    │  and Billing favour single-pane scrollable layouts.         │
│ Identity   │                                                              │
│ Roles      │                                                              │
│ Content    │                                                              │
│ Cohorts    │                                                              │
│ Audit      │                                                              │
│ Compliance │                                                              │
│ Integratns │                                                              │
│ Billing    │                                                              │
│ Branding   │                                                              │
│ Settings   │                                                              │
└────────────┴─────────────────────────────────────────────────────────────┘

E.25.2 Members workspace (canonical)#

Most-used surface. Optimise for it.

Layout (Template A):

  • Navigator (left, 240px):

    • Filter chips: All · Active · Invited · Suspended · Pending review.
    • Search input (cmd-/).
    • Group filter (multi-select chips).
    • "Bulk import" button (CSV / OneRoster) at bottom.
  • Detail (centre, fluid): Member list, virtualised. Each row:

    text
    [avatar] Name (body-md) · email (body-sm fog-300)
                                      · role pill · status pill · last sign-in · ⋯
    

    Multi-select with checkbox column; bulk action bar appears at top when ≥1 selected.

  • Inspector (right, 320px): Selected-member detail.

    • Summary card: avatar, name, email, role, group(s), status.
    • Tabs: Activity · Sessions · Devices · Audit · Permissions.
    • Inline actions: Reset password · Send sign-in link · Suspend · Remove.

Add member flow.

  • "Add member" button top-right of detail pane → modal with steps:
    1. Email (single or bulk paste, comma/newline separated).
    2. Role (default: Member; pick from tenant role catalogue).
    3. Groups (optional multi-select).
    4. Send invitation? (yes/no, with quiet-mode for SCIM-provisioned tenants).
    5. Confirmation summary, send.

Edge states: member without name (show email as primary), pending invite expired (amber chip + "Resend"), suspended member (greyed row + suspend reason chip), SCIM-provisioned member (lock icon — "Managed by IdP").

E.25.3 Identity workspace (SSO + SCIM + OneRoster)#

Three connected sub-tabs:

  • SSO — Provider selection (SAML / OIDC / Custom), metadata exchange (XML upload or manual fields), test sign-in flow with status indicator, fallback password sign-in toggle.
  • SCIM — Endpoint URL + bearer token (rotatable), supported attribute map, provisioning health (last-sync timestamp, success/error count, drift detection between SCIM source and Oshun roster).
  • OneRoster (education tenants) — district/school hierarchy import, class roster sync schedule, term/period mapping, per-import diagnostic log.

Health rail. Each sub-tab shows a coloured health pill: Healthy (success-500 dot) / Degraded (amber-500) / Failing (danger-500), with last- checked timestamp and "Run health check" button.

E.25.4 Compliance workspace#

Layout: single-pane scrollable.

Sections:

  1. Posture dashboard — GDPR · CCPA · COPPA · FERPA chips with status (✓ compliant · ⚠ action needed · — n/a).
  2. Data subject requests — Inbox of export/delete/access requests. Each row: requester, type, age, SLA, status, action.
  3. Retention policy — Per-data-type retention rules (e.g., session logs 90d, journal entries indefinite unless deleted).
  4. Consent records — User consent history; opt-in/out audit trail.
  5. Compliance evidence export — Generate signed bundle (PDF/JSON) for auditor; bundle scope picker, export history.

Anti-goals. No alarmist red banners. No badges of compliance theatre. State should be quiet, evidentiary, exportable.


E.26 Bellona Render Queue (representative pipeline workspace)#

Why specify this one explicitly. Bellona Render Queue is one of the most- used Studio operations workspaces and a useful exemplar for the "queue-with-detail" pattern reused across Bellona, Aja, Isis, and Yemaya.

Goal. Operator triages render jobs across projects: kill stuck jobs, re-prioritise queue, reroute to different render farm, debug failures.

Layout (Template A):

  • Navigator (left, 280px):

    • Filter chips: status (queued · running · paused · failed · done), type (preview · final · batch · client-review), priority.
    • Project selector.
    • Owner / submitter filter.
    • Time window (today · 24h · 7d · custom).
    • Saved views: "My queue," "Failures," "P0 backlog."
  • Detail (centre, fluid): Queue list, virtualised. Row pattern:

    text
    [priority dot] [type icon] Job ID · project · submitter · age · runtime · GPU class · status pill · ETA · render farm node · ⋯
    

    Status pulse only on Failed rows (200ms easeInOut). Sticky header. Row density 36px (compact) or 44px (comfortable).

  • Inspector (right, 360px, resizable): Selected job detail.

    • Header: Job ID, project, submitter, age, status, priority. Action bar: cancel · retry · re-prioritise · reroute · download outputs.
    • Tabs: Overview · Inputs · Outputs · Logs · Manifest · History.
    • Overview tab: pipeline graph (mini, click to expand), elapsed breakdown by stage, GPU/CPU/memory chart.
    • Inputs tab: source assets list with sizes, hashes, fetch status.
    • Outputs tab: rendered frames thumbnail strip with scrubber; per- frame inspector (frame number, render time, anomaly flags).
    • Logs tab: tail of log lines, level filter (info/warn/error), search, pause stream toggle.
    • Manifest tab: structured output manifest (JSON-like tree) with copy button.
    • History tab: state-change timeline.
  • Bulk action bar (appears when ≥1 selected): retry · cancel · re- prioritise · reroute · export selection.

Edge states: empty queue ("No jobs match. Try a different filter."), single job selected (full inspector), 1000+ jobs (virtualised list with "Showing 50 of 1247" footer), render farm node offline (warning banner inline above rows from that node), all jobs failing same error (banner "32 jobs failing with same error — root cause?").

Keyboard: j/k navigate rows; enter open selected; r retry; c cancel; p re-prioritise modal; ? shortcuts.


Part F — Cross-Cutting Requirements#

F.1 Accessibility (mandatory across all surfaces)#

  • WCAG 2.1 AA baseline on every surface. AAA on long-form Nisaba reading and Tara session player.
  • Body-text contrast ≥4.5:1, large-text ≥3:1, HC mode ≥7:1.
  • Focus rings: 2px aqua-500 outline (or HC equivalent) with 2px offset. Never rely on colour alone.
  • Keyboard navigation: every interactive element reachable; visible focus; skip-to-content link.
  • Screen reader: semantic landmarks, ARIA labels on icon-only buttons, live regions for status changes (e.g., "Saved").
  • Dynamic type: scale to 200% without breaking layout.
  • Reduced motion: see Part B.8.
  • Captions / audio descriptions: Tara session audio + ambient media must carry captions; opt-in audio descriptions for Nyx night-sky narration.

F.2 Performance budgets#

  • Mobile cold start: ≤2.5s on a mid-tier device.
  • Home hydration p95: ≤2.0s.
  • Primary navigation transition: ≤300ms perceived.
  • Largest Contentful Paint (web): ≤2.5s on 4G connection.
  • Cumulative Layout Shift: ≤0.1.
  • Bundle: initial JS ≤200KB gzipped (web shell).
  • Image: next/image or equivalent with srcSet; AVIF/WebP first.

F.3 Internationalisation#

  • Right-to-left support for Arabic, Hebrew (Nisaba especially).
  • Logical CSS properties (margin-inline-start, etc.).
  • Numeric and date formatting via locale.
  • Translation-ready strings — no string concatenation in UI logic.

F.4 AI disclosure copy#

When AI is producing content visible to the user, a quiet disclosure label must be present:

text
"AI-narrated"            Tara session whose voice is generative.
"AI-summarised"          Veritas summary above primary article.
"Grounded in your        Assistant message using user memory.
 recent activity"
"Drafted by assistant"   Arete coach reply.

Disclosure label is label-sm, fog-300, prefixed by a small spark icon (12px). Never hidden, never modal.

F.5 Copy / tone of voice#

  • Sentences ≤16 words in primary UI copy.
  • No exclamation marks in customer surfaces (except recovery messages expressing relief: "It's okay.").
  • Verbs favour the calm register: "Resume," not "Continue your journey!"
  • Recovery before correction: when user errs, lead with "It's okay" or acknowledge before suggesting fix.
  • Trust language: "Confidence," "Evidence," "Source" — never "Fact-checked ✓" or red banners.
  • Domain voice:
    • Tara: gentle, embodied, unhurried.
    • Arete: quiet conviction, "small actions, daily."
    • Veritas: precise, sourced, never alarmed.
    • Nyx: factual + reverent, "Saturn rises at 21:14."
    • Nisaba: scholarly, considered, citation-friendly.

F.6 Analytics + telemetry visibility#

Whenever a surface emits analytics events, designate a tiny dev-mode overlay (toggle in profile) that displays the event name and payload as it fires. This is for design QA — not a production feature — but useful in mockups.

F.7 Haptics (mobile)#

text
Subtle tap        Tab change, save, light confirmation
Medium tap        Primary CTA submit, session start
Strong tap        Destructive confirm, error
Selection        Picker scroll, segmented control change
Success         Daily check-in complete, milestone
None            Ambient fields, decorative motion

F.8 PWA / offline behaviour (web)#

  • Installable with manifest, icons (1024² maskable), splash, theme colour #040B16.
  • Offline shell: Home, Activity, Library, Profile rendered; saved items available; cached recent reads available.
  • Update flow: in-app banner "OSHUN updated — refresh to apply" with Refresh button.
  • Service worker: stale-while-revalidate for non-critical content; network-first for live data.

F.9 Layout grid and max-widths#

Per docs/releases/v1/scope/grid-gutter-max-width-rules.md:

  • Max content width: 1280px reading shell, 1440px admin shell, 720px long-form reading, 62ch Nisaba reading frame.
  • Gutters: 16 (mobile), 24 (tablet), 32 (desktop).
  • Columns: 4 (mobile), 8 (tablet), 12 (desktop).

Part G — Output / Deliverables Expected from Claude Design#

For each surface in Part E, request these outputs:

  1. Interactive HTML/CSS/JS prototype in Claude Design canvas.
  2. Three viewport sizes: 360 (mobile), 768 (tablet), 1280 (desktop).
  3. Three theme states: dark (default), light, dark-high-contrast.
  4. All edge states: loading, empty, error, offline, partial-failure, reduced-motion (where applicable).
  5. A tokens.css file consistent across surfaces, generated from Part B.
  6. A component inventory listing reusable components with names matching libs/oshun/ui conventions.
  7. A handoff bundle suitable for Claude Code: HTML/CSS/JS + tokens.

When you finish a surface, ask Claude Design to:

  • Verify it works at 360 width without horizontal scroll.
  • Verify keyboard tab order traverses CTA → secondary → tertiary.
  • Verify reduced-motion freezes animation.
  • Verify all images carry alt text or role="presentation" if decorative.

Part H — Anti-Goals (consolidated, paste verbatim)#

text
NOT:
- Generic SaaS card-grid homepages.
- Wellness-app clichés (pastel blobs, lotus icons, inspirational
  filler copy, "namaste" tone).
- Glassmorphism, excessive blur, metallic gimmicks.
- Per-domain visual fragmentation that breaks the one-shell model.
- Multiple competing primary actions per viewport.
- Decorative motion that competes with reading or orientation.
- Punitive streak logic, guilt copy, badge proliferation.
- Top-corner placement of session-critical mobile actions.
- Hero gradients pretending to be visually rich. Heroes need a real
  anchor.
- Centred hero above a wall of cards.
- Big coloured marketing CTA buttons in product chrome.
- Red-dot notification anxiety designs.
- "BREAKING" newsroom chyrons in Veritas.
- HUD-style overlays in Nyx.
- Trophy / podium / XP gamification in Arete.
- Skeumorphic book illustrations or fantasy parchment in Nisaba.
- Anthropomorphic AI avatars with animated faces.
- "Trusted by"-logo cloud in marketing.
- Animated number counters.
- Auto-playing audio in ambient media.
- Always-on listening indicators without explicit user action.

Part I — Quick-Paste Prompts#

These are pre-composed prompts ready to paste into Claude Design. Each assumes Part B has already been pasted as the system primer.

I.1 Quick prompt: Welcome / public landing#

text
Generate the OSHUN public welcome page.

Goal: First impression of OSHUN as a quiet observatory for inner life,
opening into five rooms (Tara, Arete, Veritas, Nyx, Nisaba). One anchor
hero, no SaaS card grid, premium through restraint.

Audience: Cold visitor referred from press or word-of-mouth.

Hero anchor: Pick a single dominant composition — full-bleed nocturnal
photograph with a constrained text column over a calm tonal area. Add a
subtle controlled ambient parallax (slow stars, low amplitude). Provide a
still-frame fallback.

Layout (top to bottom):
1. Hero: anchor image + one-sentence kicker (label-md, fog-300) +
   display-xl headline + body-md sub + single aqua-500 CTA "Open OSHUN".
2. Five-room strip: five cards, accent edge top in domain colour, icon,
   name, single-line promise. Mobile: vertical stack.
3. One-paragraph thesis using display-lg opener.
4. Three "moments" panels (morning, midday, evening), each with one image.
5. Quiet trust strip — small citations, no logo cloud.
6. Minimal footer ink-950 background.

Tokens: use the OSHUN foundation. ink-950 #040B16, aqua-500 #20C9D8, domain
accents Tara #2BC6B4, Arete #3AC77B, Veritas #27A3F3, Nyx #7D7CFF, Nisaba
#C88B2C. Type: Space Grotesk display, Manrope body.

Deliver: dark default + light + dark-HC themes; viewports 360, 768, 1280.
Reduced-motion freezes ambient field. Hero must work as still frame at
360. WCAG AA on all text over media.

Anti-goals: gradient-only hero, card grid above the fold, "trusted-by" logo
cloud, animated number counters, wellness-blob illustrations.

I.2 Quick prompt: Customer Home (web/PWA)#

text
Generate the OSHUN customer Home page (web/PWA).

Goal: User sees time-of-day, continuity card, one-tap primary domain action,
five rooms, recommendations. Calm, scannable, two-tap-to-value.

Audience: Daily user (any P1 persona). 80% of sessions land here.

Layout (1280 baseline, three breakpoints):

Top bar (56px): wordmark · search input · notifications icon · avatar.
Left sidebar (240px): Home · Explore · Activity · Library · Profile · five
domain shortcuts.
Main content:
- Greeting kicker (label-md): "Good evening, [Name]"
- Anchor headline (display-lg): "Quiet evening for you."
- Subline (body-md, fog-300): time/state context.
- Continuity card: full-width raised surface, "Continue: 7-min Calm Drift
  (Tara)" + Resume button (aqua-500 primary) + Dismiss/Switch text actions.
- Quick-action chip row (4 chips, aqua-500 outline on hover): "3-min
  breath", "Today's claim", "Tonight's sky", "One habit."
- Five-room strip: 5 cards across with domain accent top edge (4px), icon,
  name (title-md), continue-state (body-sm), inline CTA.
- Recommendations rail: 3 cards, each with "why this" chip explaining
  recommendation in ≤6 words.

Tokens: ink-950 canvas with subtle vertical gradient #08152A → #0A2036 →
#040B16. Cards: ink-900 with 1px border at 8% white. Domain accents per
foundation.

Edge states required:
- Loading skeleton with restrained shimmer.
- First-time empty state: continuity card replaced by "First moments"
  card; recommendations replaced by domain primer carousel.
- Partial-failure: domain unavailable shows quiet "Tara unavailable" inside
  its card; rest of shell live.
- Reduced-motion: no card stagger, instant load.

Mobile breakpoint (414): sidebar becomes bottom nav (5 tabs); plus
bottom-left FAB (Domain Switcher) and bottom-right FAB (Assistant).

Anti-goals: no metric-chip dashboard, no animated number counters, no
"trending" lists, no second sidebar.

I.3 Quick prompt: Tara Session Player#

text
Generate the OSHUN Tara session player screen (mobile + web).

Goal: User starts a meditation in ≤2 taps from Tara entry. Screen should
settle the body before asking anything.

Layout: Full-screen. Top 70%: ambient still or low-amplitude video field
(slow gradient cloud, water surface, or grounded photograph). Lower 30%:
dim band on ink-900 with semi-transparent overlay.

Lower band content:
- Session title (title-md), centred.
- Instructor name (body-sm, fog-300).
- AI-disclosure label if AI-narrated ("AI-narrated", label-sm with spark
  icon).
- Scrub bar (1px height, domain-tara #2BC6B4 progress fill).
- Time labels: elapsed / total (mono-sm).
- Transport controls: skip-15s, play/pause (large 64px circle, aqua-500
  surface, ink-950 icon), skip-15s.
- Bottom-corner exit X.

Pre-session state: 8-second grounding moment with breath-pulse ring
(4-7-8 cadence), large display-lg copy "Take a breath", optional skip CTA.

Post-session state: "How did that feel?" with three quiet chips: calmer,
same, restless. Save & return to Tara Home.

Reduced-motion: ambient field freezes; breath ring replaced by text cue
"Inhale 4 · Hold 7 · Exhale 8".

Anti-goals: no emoji feeling pickers, no "session #N", no streak
celebration, no spinning loaders during play.

I.4 Quick prompt: Veritas Story Reader with Evidence Rail#

text
Generate the OSHUN Veritas story reader (web + mobile).

Goal: User reads a verified story confidently. Evidence is visible without
making the surface feel clinical.

Layout (desktop 1280):
- Top bar with breadcrumb (Veritas › Topic › Story).
- Reading frame, 720px max width, 62ch measure, body-lg 17/28, paragraphs
  spaced 24px.
- Persistent right rail (320px) "Evidence":
  - Confidence chip at top: "[High confidence ●]" success-500 dot,
    label-md.
  - Source list: each row has source name, credibility tier (label-sm),
    publication date, link icon.
  - Inline citation jump-to: small numbered pills [1] [2] [3] inline in
    body text; clicking scrolls and highlights source in rail.
- On mobile (414): evidence rail becomes a bottom drawer; pill citations
  open it.

Visual style:
- domain-veritas #27A3F3 on confidence dots, evidence rail edge, citation
  pills.
- Pull quotes in serif, display-sm, indented 10%.

Edge states:
- Contested confidence: amber-500 dot, paired with "See contested" toggle
  that expands counter-evidence.
- Unverified: fog-300 dot with "We're still verifying" inline note.
- Source unavailable: fog-300 row, italic, "Source archived."

Anti-goals: no red BREAKING chyron, no fear icons, no partisan colour
coding, no autoplay video.

I.5 Quick prompt: Nyx Tonight's Sky#

text
Generate the OSHUN Nyx "Tonight's Sky" home view (mobile + web), with
night-mode variant.

Goal: User in evening mode opens Nyx and immediately sees what's happening
tonight; can transition to interactive sky map in one tap.

Layout (mobile 414, primary use case):
- Hero: full-bleed slow ambient star canvas (low-amplitude particle motion,
  ≤0.02px/frame), ink-950 base; text overlay sits in a calm darkened
  region.
- Hero text: kicker label-md "Tonight, near you", display-lg "Saturn rises
  at 21:14", body-md sub describing visibility.
- Primary CTA: "Open sky map" (aqua-500, ≥56px tall, primary zone).
- Below hero: "Three things to watch tonight" — 3 quiet rows with object
  name, rise time, magnitude.
- Below: "This week's events" — calendar strip.
- Below: recently viewed objects.

Night mode (auto-engages after sunset based on local time):
- Switch ink-950 to a deep red-tinted near-black (#0A0202).
- Replace aqua-500 accents with warm amber-red (#E89B6B).
- Reduce overall luminance.
- Disable bright-white text; switch to a desaturated warm grey.

Web version:
- Hero takes 60vh; below: 3-column grid of night-events, this-week,
  recently-viewed.

Anti-goals: no neon, no HUD, no rocket emojis, no generic space wallpaper.
The sky photograph or canvas must be specific and credible.

I.6 Quick prompt: Arete Daily Check-in#

text
Generate the OSHUN Arete daily check-in screen (mobile + web).

Goal: ≤60s daily check-in for a power user. Three quiet steps, no
gamification.

Layout: vertical, three card sections, scrollable on mobile.

Section 1 — "Yesterday" (recovery-aware):
- Title: "How did yesterday go?" (title-md)
- Three chips: "Good", "Mixed", "Tough" (label-md, no emoji).
- Optional one-line note input.

Section 2 — "Today's intention":
- Title: "One thing for today?" (title-md)
- Single-line input, body-lg (placeholder "Write a small step").

Section 3 — "Anything to log?":
- Optional habit tick row (3 habits as toggleable chips with current
  streak compactly shown next to each).
- Mood chip row (5 calm chips: settled, focused, restless, low, anxious).

Submit button (aqua-500): "Save check-in" (≥56px on mobile).

After missed day: section 1 leads with body-md "It's okay. Pick today
up." (success-500 leading), no streak counter visible until user opens
streak settings.

Domain accent: domain-arete #3AC77B replaces aqua on Submit and on
today-completed states inside Arete chrome only.

Anti-goals: no leaderboards, no XP, no animated trophies, no "you broke
your streak" panic copy.

I.7 Quick prompt: Nisaba Reading Desk#

text
Generate the OSHUN Nisaba long-form reading desk (web + mobile).

Goal: editorial, durable surface for an hour-long reading session.

Layout (desktop 1280, three panes):
- Left pane (compare slip, 280px, optional): a second passage variant
  (translation, edition); collapsible.
- Centre pane (reading frame, 720px max, 62ch measure):
  - Passage title (display-md).
  - Author/edition meta (label-sm, fog-300).
  - Body in body-lg, 1.6 line-height, 24px paragraph spacing,
    left-aligned, never full-justified.
  - Drop-cap optional on first paragraph.
  - Pull quotes serif, display-sm, indented 10%.
  - Numbered footnote markers inline; on hover/tap reveals lexicon panel.
- Right pane (notes / lexicon, 320px):
  - Tabs: Notes · Lexicon · Concepts.
  - Notes: thread of user annotations attached to passage anchors.
  - Lexicon: when a word is hovered/tapped, panel shows morphology,
    etymology, frequency, related passages.
  - Concepts: links to concept graph nodes referenced in passage.

On mobile (414): single-pane reader; left and right panes become drawers
accessible via top-bar icons.

Domain accent: domain-nisaba #C88B2C on highlight pin, lexicon panel
edge, footnote markers.

Reading-mode controls (top-bar): font-size +/-, line-height, theme (dark
default · light · sepia for AAA-aspiring reading), focus mode (hides UI
chrome).

Edge states:
- AAA reading mode: light theme with #FBF7EE background, ink-950 text,
  contrast ratio ≥7:1, line-height 1.7, full-screen.
- Reduced-motion: no lexicon panel slide-in; appears instantly.
- Long passage: scroll progress thin bar at top, no infinite scroll.

Anti-goals: no skeumorphic book images, no fantasy parchment textures,
no reading-streak gamification, no animated page-turn effects (calmer to
just scroll).

I.8 Quick prompt: Studio Review/Approval Workflows workspace#

text
Generate the OSHUN Studio "Review and Approval Workflows" workspace.

This is a Studio workspace, not a customer surface. Visual posture is
operator-grade, dense, low-noise, keyboard-first — "an air-traffic control
room for governed intelligence work." Reference: Linear, Stripe Dashboard.
Do NOT use customer canvas (no nocturnal gradient, no calm hero, no domain
accents on chrome).

Goal: Reviewer triages submitted review packages, opens a package, sees
context + evidence + recommendation, decides Approve / Deny / Defer / Request
Changes. Operator should be able to clear a 30-package queue in one focused
session.

Personas: Editor, Reviewer, Approver, Auditor.

Layout (Template A — three-pane):

Top bar (40px): org/tenant switcher · cmd-K · user.

Workspace header (56px sticky):
  "Review & Approval Workflows" · [tenant chip] · [release chip]
  Action bar (right-aligned): Bulk approve · Bulk deny · Export evidence.

Tabs: Overview · Journeys · Evidence · Audit · Settings.

Default tab = Overview. The Overview canvas uses the three-pane Template A:

Left navigator (240px):
  - Filter chips: All · Pending · In review · Approved · Denied · Deferred ·
    Needs changes.
  - Stage filter (multi-select chips: Stage 1 / Stage 2 / Final).
  - Submitter filter.
  - Age slider.
  - Saved views: "Mine," "Aged > 24h," "P0."

Centre detail (fluid):
  Virtualised list of review packages. Row pattern:
    [priority dot] [type icon] Package title (body-md) · submitter · stage ·
    age · SLA timer pill · assignee · status pill · ⋯

  Sticky header, sortable columns, row density toggle (36px compact /
  44px comfortable). Multi-select checkbox column; bulk action bar appears
  when ≥1 selected.

Right inspector (360px, resizable):
  Selected package context. Tabs:
    - Summary: title, submitter, stage, current decision, recommendation,
      lineage chip, blocking dependencies.
    - Evidence: structured evidence rows (validation runs, policy checks,
      diff vs baseline, citation links).
    - Decision: action buttons (Approve / Deny / Defer / Request Changes),
      decision rationale composer (markdown), routing target picker.
    - Discussion: comment thread with @mentions, file attachments inline.
    - Audit: append-only state-change timeline.

Tokens: ink-900 canvas (no gradient). title-lg max for headings. Severity
pills: S1 danger-500, S2 amber-500, S3 fog-300. Status pills: dot only,
no fill wash. Decision chips: outline only (success-500 / danger-500 /
fog-300). Diff: removed = danger-500/12% bg, added = success-500/12% bg.

Keyboard: j/k navigate rows; enter open; a approve; d deny; r request
changes; f defer; ? shortcuts.

Edge states:
- Empty queue: "No packages match. Try a different filter."
- Validation failing: inline pill on row "Validation failed" (danger-500).
- Stale package (age > SLA): row left-edge red 2px stripe + breach pulse.
- Approval requires step-up auth: confirm modal with biometric/PIN prompt.

Anti-goals: no customer hero, no card-grid mosaic, no permanent red
atmosphere, no decorative motion, no domain accents on chrome.

I.9 Quick prompt: Tenant Console Members workspace#

text
Generate the OSHUN Tenant Console Members workspace (institutional
admin surface).

Visual posture sits between Studio's operator density and the customer
shell's calm: lighter than Studio, denser than customer. Tenant branding
is allowed (logo + accent colour) within the Oshun design grammar; the
overall chrome remains Oshun.

Goal: Tenant admin manages member roster — search, filter, add, suspend,
remove, change role, bulk import via CSV/OneRoster, audit per-member.

Personas: Tenant Administrator, Identity Operator, Compliance Auditor.

Layout (Template A — three-pane):

Top bar (44px): [Tenant logo] Tenant Name · workspace search · cmd-K ·
admin avatar.

Workspace header (56px sticky):
  "Members" · [filter summary chip e.g., "Active · 247 of 312"]
  Action bar: Add member · Bulk import · Export.

Left navigator (240px):
  - Status chips: All · Active · Invited · Suspended · Pending review.
  - Search input (cmd-/).
  - Group filter (multi-select chips).
  - Role filter.
  - Bulk import button at bottom (CSV / OneRoster).

Centre detail (fluid):
  Virtualised member list. Row pattern:
    [avatar 28px] Name (body-md) · email (body-sm fog-300) · role pill ·
    group chips (max 2 visible, +N for overflow) · status pill ·
    last sign-in (label-sm) · ⋯

  Sticky header. Sortable columns: Name, Role, Last sign-in. Multi-select
  checkbox column. Density toggle 36/44px.

  Bulk action bar (appears with ≥1 selected): Change role · Add to group ·
  Send invitation · Suspend · Remove · Export selection.

Right inspector (320px):
  Selected-member detail.
  - Summary card: avatar (large), name, email, role, group(s), status,
    "Managed by IdP" lock icon if SCIM-provisioned.
  - Tabs: Activity · Sessions · Devices · Audit · Permissions.
  - Inline actions: Reset password · Send sign-in link · Suspend · Remove.

Add member flow (modal):
  Step 1: Email (single or bulk paste, comma/newline separated).
  Step 2: Role picker (from tenant role catalogue, default: Member).
  Step 3: Groups (multi-select).
  Step 4: Send invitation? toggle (off for SCIM-provisioned tenants).
  Step 5: Confirmation summary, primary "Send" CTA (aqua-500).

Tokens: surface-base #F7F8FA (lighter than Studio's ink-900 — this is a
tenant-branded admin surface that should feel approachable but
operator-grade). text-primary ink-950, text-secondary fog-700. Tenant
accent appears only on logo and on a thin top-bar underline (2px); no
chrome-wide tenant tinting.

Edge states:
- SCIM-provisioned member: lock icon + tooltip "Managed by [IdP name]";
  Edit role disabled.
- Pending invite expired: amber-500 status pill + "Resend" inline action.
- Suspended member: row greyed; suspend reason chip on hover.
- Bulk import in progress: top inline progress bar showing N of M
  processed; live diagnostic log accessible.
- Empty roster: editorial empty state with one CTA "Add your first
  member" or "Connect identity provider."

Keyboard: j/k navigate rows; enter open inspector; a add member; b bulk
import; / search; ? shortcuts.

Anti-goals: no consumer hero, no marketing CTA buttons, no badges-of-
compliance theatre, no chrome-wide tenant brand wash, no toggle switches
that look gamified.

I.10 Quick prompt: Admin Web Inbox#

text
Generate the OSHUN Admin inbox.

Goal: Operator triage of cross-workspace items (queue items, incidents,
reviews, notifications). Calm, dense, professional.

Layout (1440 baseline):
- Left rail: workspace navigation (Inbox, Queues, Incidents, Governance,
  Tenants, Analytics, Settings).
- Top bar: organisation switcher · search · operator avatar.
- Main content split:
  - Filter row: source (queue/incident/review), priority chips, age,
    assignee, status.
  - Item list (60% width): rows with [priority dot] [type icon] Title ·
    submitter · age · SLA timer pill · assignee · row actions.
  - Detail pane (40% width): selected item context, action buttons, audit
    trail, related links.

Visual style:
- Canvas: ink-900 (no nocturnal gradient).
- Headings: title-lg max (no display-lg in admin).
- Density: rows 44–48px; padding tighter than customer.
- SLA timer pill: time remaining label-sm; green ≥30%, amber 10–30%, red
  <10%; pulse only on red, 200ms interval, easeInOut.
- Severity badge: S1 danger-500, S2 amber-500, S3 fog-300, label-sm,
  weight 600.

Anti-goals: no customer "calm hero", no domain accents on shell chrome,
no nocturnal gradient, no decorative motion. Density is fine; anxious
density is not.

Edge states: loading skeleton list rows; empty inbox "Nothing to triage";
single item selected vs multi-select bulk action mode.

Part J — Glossary#

text
Shell            The shared customer chrome (top bar, sidebar/tabs,
                 navigation, search, notifications, assistant) that
                 hosts all five domains.

Room             A domain (Tara, Arete, Veritas, Nyx, Nisaba) presented
                 inside the shell. "One shell, five rooms."

Continuity       The user's most likely next action across domains.
                 Surfaced on Home as the continuity card.

Domain accent    The single hex token assigned to each domain. Tints
                 local rails, focus areas, progress, emphasis. Never
                 full-screen.

Assistant        The global AI conversation overlay accessible from
                 any surface.

Memory           User activity, preferences, notes that the assistant
                 can ground replies in. Disclosure required.

Evidence rail    Veritas-specific persistent panel showing confidence
                 chip, sources, citation jump-to.

Reading frame    Nisaba-specific 62ch composed body region with
                 specific line-height, paragraph spacing, justification
                 rules.

V1 foundation    The token bundle in `libs/oshun/design-tokens/` that
                 all surfaces consume.

Two-taps-to-value Contract that primary domain action is reachable in
                 ≤2 taps from Home for every P1 persona.

Part K — Source Documents (Repository)#

If Claude Design has codebase access, point it to these for ground truth:

text
docs/releases/v1/design/brand-system.md
docs/releases/v1/design/ux-principles.md
docs/releases/v1/specs/personas.md
docs/releases/v1/specs/product-brief.md
docs/releases/v1/scope/mvp-scope.md
docs/releases/v1/design/motion-language.md
docs/releases/v1/design/ergonomics-requirements.md
docs/releases/v1/design/haptics-guidelines.md
docs/releases/v1/design/readability-standards.md
docs/releases/v1/design/customer-visual-thesis.md
docs/releases/v1/design/admin-visual-thesis.md
docs/releases/v1/scope/imagery-illustration-texture-ambient-media-rules.md
docs/releases/v1/scope/immersive-vs-utilitarian-posture-rules.md
docs/releases/v1/design/ai-disclosure-copy-guidance.md


docs/releases/v1/scope/partial-failure-and-degraded-mode-rules.md
docs/releases/v1/specs/program-charter.md
docs/releases/v1/specs/web-app-admin-v1-prd.md
libs/oshun/design-tokens/src/tokens.ts
libs/oshun/ui/src/
apps/oshun/web/src/design-system/
apps/oshun/mobile/src/motion/mobileMotion.ts

Part L — Acceptance Review (paste at end of every Claude Design session)#

Before accepting any generated surface, verify:

  • Does the first viewport reduce noise instead of adding stimulation?
  • Is there one dominant action or takeaway per viewport?
  • Does domain identity appear as atmosphere, not chrome?
  • If brand accents were reduced 20%, would the screen still feel premium and unmistakably OSHUN?
  • Does the surface feel designed around the user's body (thumb zone, reading flow, return path)?
  • Are loading, empty, error, offline, partial-failure states present?
  • Reduced-motion variant works?
  • Dark, light, and high-contrast themes all rendered?
  • WCAG AA contrast on all text?
  • Tap targets ≥44px on mobile?
  • Keyboard tab order traverses primary → secondary → tertiary?
  • All AI-generated content carries disclosure label?
  • No anti-goal violations (Part H)?

If any answer is no, iterate.