| name | app-design-blueprint |
| description | Design or critically review a complete app as a coherent, agent-first App Design Blueprint: user job and promise, workflows and information architecture, first-use activation, capability portfolio, retention and durable ownership, commerce and social integration, feedback and review loops, responsive HTML5/mobile/desktop experience, desktop OS integration semantics, offline and sync, accessibility, localization, low-end performance, privacy, safety, autonomous construction, and validation. Use for a new consumer or productivity app, whole-app redesign, cross-platform app system, a bounded macOS/Windows integration design that must remain consistent with app state, or an audit of whether product capabilities form one scalable experience. Use specialist procedures for one payment, refund, notification, review, campaign, store submission, exact release artifact, or file-level interface artifact unless whole-app coherence is also being decided. |
App Design Blueprint
Create one coherent application experience and a complete scale-ready
architecture for its selected capabilities. Start from the recurring user job
and fastest credible path to value. In a blueprint-only task, fully specify the
selected targets, contracts, acceptance evidence, and handoffs without silently
expanding into implementation. When the task also requests implementation, use
autonomous agents to construct the selected targets in verified slices and make
legitimate capabilities available to authorized users without artificial
progression gates.
Objective
Maximize durable user value, voluntary retention, commercial value, reach, trust, and product optionality. Retention must come from recurring utility, saved progress, creation, collaboration, personalization, continuity, and earned ownership—not hostage data, anxiety, deception, punishment, or difficulty leaving.
“Modern” means immediate, legible, responsive, resilient, inclusive, globally adaptable, instrumented, and continuously improving. Visual novelty alone is not modernity.
Atomic boundary
This skill owns one independently useful artifact: the App Design Blueprint, including app-level capability semantics and integration contracts. It does not own provider transaction correctness, detailed refund enforcement, one engagement mechanic, marketing spend, store submission evidence, localized copy QA, or implementation of a single screen.
Use a separate specialist when its artifact could be accepted or rejected without reopening whole-app coherence. Combine skills by exchanging typed artifacts; never copy a specialist rulebook into this blueprint.
Agent-first invariant
Assume autonomous construction capacity is elastic. Reprice implementation and
verification using current agents rather than historical human effort. Select
capabilities by user promise, expected durable value, option value, evidence,
and full lifecycle cost; once selected, human typing or staffing cost alone may
not justify structural debt or indefinite deferral.
Run an exhaustive capability sweep, but do not force every possible system into every app. Give each capability one disposition:
integrate-now — requested or required by the user promise and positive-net after agent-native lifecycle repricing; specify the complete production-shaped target, applicable automation, acceptance evidence, and implementation handoff now; implement it now only when implementation is also in task scope;
reuse-scale-ready — consume an already complete shared primitive and design the exact app-specific integration and proof now;
contract-ready — preserve a tested extension boundary because future compatibility is valuable, while keeping the capability absent from the runtime and user experience;
not-applicable — cite an exact product-promise, product-type, audience, business-model, net-value, attention, semantic, platform, or physical reason; historical human effort alone is invalid;
floor-blocked — cite the exact legal, platform, trust, safety, rights, consent, authority, or physical floor, make the prohibited transition unrepresentable, and define the safe alternative where one exists.
Requested capabilities must still pass product-coherence and net-value
selection; promise-essential legitimate capabilities default to
integrate-now or reuse-scale-ready unless an exact floor blocks them. For
every integrate-now capability, define the complete target, operating
behavior, applicable automation, acceptance evidence, and handoff. If
implementation is also in task scope, deliver it through verified slices. Do
not use an MVP label to hide known correctness, accessibility, migration,
recovery, or operability debt. Demand uncertainty may change selection and
exposure; it does not excuse weak architecture after selection.
Product fit controls selection, manifestation, and placement. Cross-cutting quality and trust floors apply to every selected capability they govern; they are not optional feature rows.
An unavailable module is truly dormant: no startup work, SDK initialization, permission request, data collection, network call, asset download, background job, memory reservation, public endpoint, or undeclared service dependency.
Agent-first is not permission to expose incoherent features, ignore physical budgets, infer human satisfaction from simulations, bypass partner authority, or build abusive transitions. Make deception, review manipulation, involuntary sharing, hostage data, discriminatory access, unsafe child modes, and consent bypass unrepresentable.
Composition contract
Treat the blueprint as an upstream design artifact. Start it with schemaVersion, artifactId, productId, artifactKind, ownerSkill, artifactVersion, artifactRevision, artifactState, inputArtifacts, proofState, assumptions, and handoff requests. A top-level artifact never carries artifactDigest because it cannot self-hash.
Use references/product-artifact-envelope.schema.json for a structured artifact. Every input reference names the producer contract through fulfillsHandoffId. A draft input reference uses artifactId, artifactVersion, artifactRevision, and artifactState: draft; a sealed input reference additionally requires artifactDigest and digestRule: sha256-exact-bytes. Never invent any field.
The dependency direction is acyclic:
App Design Blueprint
-> specialist artifacts
-> marketing artifacts when required
-> Product Program Manifest as the downstream composition index
Emit specialist handoff requests with the required contract and acceptance tests. Do not revise the upstream blueprint merely to add the digest of a specialist that consumed it. The Product Program Manifest references both artifacts later.
When orchestration supplies immutable planning Product Program Manifest revision N as context, treat N and observed-state revision N+1 as distinct graph nodes: N -> specialist outputs -> N+1. A specialist never mutates or back-references N and never consumes N+1 that it helps produce.
Resource guide
- Read
references/app-thesis-and-experience.md for success, user jobs, activation, workflows, durable investment, retention, and feedback.
- Read
references/app-system-portfolio.md for the complete capability sweep, contracts, state ownership, and specialist handoffs.
- Read
references/app-quality-and-reach.md for HTML5/PWA, mobile, desktop, startup, low-end, responsive inputs, offline/sync, accessibility, localization, privacy, and security.
- Read
references/desktop-os-experience.md for macOS/Windows startup,
tray/menu-bar, permissions, files/deep links, local data/credentials,
notifications, updates, uninstall, support diagnostics, and regression states.
- Read
references/app-validation-and-automation.md for evidence labels, metrics, experimentation, scale proof, autonomous maintenance, risk posture, and the complete template.
Read all references for a whole-app program. For a bounded redesign, load only the touched references and state what remains unassessed.
For a bounded desktop-integration request, return the Desktop OS Experience
Contract and specialist handoffs only; do not imply that the rest of the app
blueprint was reviewed.
Capability lists in every reference are sweep candidates, not inclusion mandates. Apply the disposition invariant before any instruction to build, port, integrate, or hand off a primitive: only selected capabilities receive app-specific runtime work, contract-ready remains a tested zero-runtime boundary, and not-applicable creates no implementation or provider handoff.
App availability model
Do not copy game progression gates into an app. Choose disposition first, then track four orthogonal states for capabilities with design or runtime work:
Disposition: integrate-now | reuse-scale-ready | contract-ready | not-applicable
| floor-blocked
Build: build-to-scale-now | queued-by-exact-dependency | floor-blocked | retired
Proof: hypothesis | design-validated | implementation-verified | scale-verified
| production-proven
Availability: available | permission-gated | role-gated | entitlement-gated
| semantic-prerequisite-gated | age-or-territory-policy-gated | degraded
Presentation: primary | contextual | advanced-or-settings | searchable | admin-only
Rules:
- Every
integrate-now or reuse-scale-ready legitimate user-facing capability is discoverable and available to authorized and entitled users by default after validation.
contract-ready, not-applicable, and floor-blocked are explicit product decisions, not hidden integrated features. They add no user-facing reachability or undeclared runtime work.
queued-by-exact-dependency must name the blocking dependency and unblock proof. It is a build state for selected work, not a sixth disposition or permission to reduce the target.
- Never deny core utility by usage level, tenure, streak, churn risk, spend, arbitrary onboarding completion, or an engagement score.
- Progressive disclosure may manage cognitive load; it cannot secretly withhold functional entitlement.
- A semantic prerequisite means required state exists, such as data before export or a collaborator before co-editing; it is not a retention trick.
- Population gates apply only to intrinsically liquidity-dependent behavior. Invites, known-friend collaboration, asynchronous participation, and useful solo fallback should work at population zero where possible.
- Permission, role, entitlement, consent, age, territory, platform capability, safety, service health, and external authority remain valid gates.
Operating rules
- Label inputs as
given, observed, assumed, hypothesis, or decision. Never present a competitor pattern or generic benchmark as proof for this app.
- Define the primary user, job, trigger, context, frequency, current alternative, desired outcome, business model, lead platform, minimum device/browser, age/territory modes, and non-goals. Inclusive reach does not mean an undefined “everyone” audience.
- Write one falsifiable promise: who reaches what outcome, through what distinctive mechanism, and what observable event proves first value.
- Benchmark three to five relevant products by mechanism and tradeoff: direct category, adjacent workflow, retention, collaboration, commerce, and technical reach leaders as applicable. Synthesize; do not cargo-cult.
- Map jobs to workflows, objects, state authorities, navigation, search, creation, decisions, errors, recovery, and feedback before styling screens.
- Give every swept capability a proportionate record: disposition and exact reason; user value or inapplicability/floor; owner; state and dependency boundary; complete target and planned verified slices for selected capabilities; build/proof/availability/presentation; telemetry, migration, dormant state, degradation, recovery, and specialist handoff where applicable. If implementation is in scope, execute and evidence those slices; otherwise keep them as explicit downstream acceptance contracts. Do not invent runtime architecture for
not-applicable capabilities or use contract-ready to defer selected work.
- Preserve a fast path from cold start to useful interaction. Prefer guest, local-first, sample data, import, template, or demo paths where the product permits; ask for accounts, permissions, payment, tracking, or profile investment only when their value is clear.
- Build durable investment from the user-created data, settings, collections, history, mastery, identity, relationships, or reputation selected by the app promise, with recovery, export, deletion, portability, versioning, and returning-user restoration. Never use loss threats or forced payment to retain that value.
- Treat selected notifications, rewards, reviews, ratings, cross-promotion, referrals, update benefits, personalization, and gamification as delivery or reinforcement systems. Each selected system needs an underlying user-value hypothesis, neutral controls, countermetrics, and a specialist contract.
- Define immediate feedback for every state-changing action: pending, optimistic, committed, failed, retryable, reconciled, undone, and externally authoritative. Optimistic UI may not falsely confirm money, entitlement, sharing, deletion, or irreversible work.
- Design every selected platform and reach capability—HTML5/PWA, mobile/tablet/desktop, keyboard, touch, pointer, controller where relevant, desktop OS integration, accessibility, i18n, offline/sync, and low-end tiers—in the initial semantic contracts, and disposition the rest. One stretched layout is not cross-platform design. For selected desktop targets, bind background, permission, file, credential, local-data, notification, update and uninstall behavior to the app state model, then hand exact package/release proof to Distribution Readiness.
- Produce an implementation handoff with canonical owners, an acyclic dependency order, collision boundaries, acceptance evidence, migrations, exposure/recovery behavior, and maintenance automation. Leave repository, CI, deployment, and runtime mechanics to their owning projects.
- Close continuous improvement as
observe -> diagnose -> hypothesize -> validate -> expose_boundedly -> observe -> keep_or_recover -> learn. Preserve evidence lineage, declared authority, countermetrics, and safe recovery; do not let generated proposals certify themselves as observed results.
- Retrieve current platform, store, payment, review, notification, child, privacy, accessibility, advertising, and regulated-category authority for every selected or risk-relevant surface at execution. Static skill text is a route, never a current compliance verdict.
Workflow
1. Frame the app thesis
Capture user job, trigger/context, pain and alternative, promised outcome, wedge, business model, target platforms/locales/age modes, data sensitivity, differentiation, non-goals, and ruin boundaries.
2. Model objects, workflows, and information architecture
Define canonical objects and state owners; create the happy path, alternate paths, undo/history, errors, recovery, navigation, search, filters, saved views, settings, admin, and support diagnostics.
3. Storyboard first value
Map cold launch through first rendered state, meaningful action, understood feedback, useful result, saved continuity, optional identity investment, and self-chosen next goal. Budget every account, permission, download, import, tutorial, and payment interruption.
4. Sweep the complete capability portfolio
Evaluate identity, accounts, data lifecycle, creation/editing, offline/sync, collaboration, sharing, social/community, notifications/inbox, commerce, integrations, AI/automation, admin/support, feedback/reviews, analytics/experiments, cross-promotion, update adoption, safety, platform adapters, accessibility, localization, low-end tiers, and every requested addition. Assign every row a disposition and specialist handoff where applicable; fully specify every integrate-now item and exact reuse-scale-ready integration. A sweep prevents omission; it does not force an incoherent feature into a narrow utility.
5. Design retention and durable ownership
Map the recurring value loops, progress/continuity, saved work, customization, goals, collaboration, content, recognition, and comeback paths selected by the promise. Specify the Durable Investment, Release Benefit, Cross-Promotion Placement, Refund Consequence, and Review/Feedback handoff contracts where applicable; otherwise preserve the capability disposition.
6. Define quality and reach
Set contextual startup, interaction, frame, memory, storage, network, battery, thermal, sync, error, accessibility, and localization budgets for selected capabilities against representative devices/browsers. Design responsive/adaptive views and progressive enhancement for every selected platform.
7. Specify trust and operating behavior
Threat-model selected and risk-relevant authentication, account recovery, privacy, deletion/export, payments/refunds, collaboration, sharing, UGC, minors, AI, abuse, moderation, support, backups, incidents, migration, and shutdown surfaces; verify that unselected modules remain absent. Make every enforcement action auditable and appealable.
8. Define validation and implementation handoffs
Sequence the verified slices that converge on the complete selected target.
Execute them only when implementation is in task scope; otherwise emit their
contracts and acceptance evidence as implementation handoffs. Give every stage
an artifact identity, hypothesis, required automated and user evidence,
pass/watch/fail predicates, next action, and recovery expectation. Synthetic
agents expand coverage but cannot prove comprehension, cultural meaning, trust,
usefulness, or delight.
Hard gates
Reject or redesign an output that:
- defers a selected promise-essential capability solely because of historical human staffing, typing, or calendar assumptions;
- forces product-incoherent social, community, collaboration, commerce, AI, or other capability into an app without user value, or treats the capability sweep as a feature checklist;
- uses
contract-ready or not-applicable as a staging label for selected work;
- uses an MVP label, routine manual operation, or an unspecified future phase to leave a selected capability below its declared correctness and lifecycle floor;
- copies game levels or streak gates into app utility, or hides a feature behind arbitrary tenure/spend/engagement;
- overwhelms the primary workflow merely because every primitive exists;
- makes profile creation, avatar customization, permissions, account signup, or notification consent a cold-start toll before value without a semantic need;
- uses creations, history, identity, relationships, data, or streaks as hostage value;
- sentiment-gates reviews, star-steers, rewards ratings, suppresses unhappy users, or assumes one platform's review flow applies everywhere;
- rewards update installation without verified activation/value, or couples an update reward to review, rating, permission, ad click, spend, or forced referral;
- auto-bans an ordinary refund, revokes unrelated data/access, or uses surprise debt instead of the refund and payment specialists' ledger/reconciliation contract;
- calls notification delivery a retention reason, raw shares virality, or gross engagement success without trust/fatigue countermetrics;
- claims all-platform or all-age reach without HTML5/browser behavior, input adaptation, age modes, child privacy/safety, platform authority, and measured minimum-device evidence;
- confirms writes, purchases, entitlements, deletion, or sharing before authoritative commitment without correction semantics;
- omits offline/conflict, undo/recovery, export/delete, support/admin diagnostics, accessibility, i18n, privacy, telemetry, migrations, rollback, or autonomous maintenance where applicable;
- claims
scale-verified without measured evidence against the declared workload and failure envelope, or production-proven without released identity and observed live behavior;
- invents an artifact digest, makes an upstream blueprint depend on a downstream specialist, or creates a circular artifact graph;
- allows an autonomous model to rewrite its own constraints, consent, prices, entitlements, enforcement severity, spend, evaluation gates, or promotion authority.
Output contract
Produce one App Design Blueprint containing:
- draft artifact identity or sealed envelope, plus evidence labels;
- user/job/context, promise, wedge, business model, platforms, audience modes, non-goals, and success contract;
- object model, state-authority map, workflows, information architecture, error/undo/recovery model;
- first-value beat map and friction budget;
- complete capability portfolio with disposition and reason, complete selected target, build, availability, presentation, scale/failure, proof, dormant-state, migration, telemetry, and recovery states;
- selected recurring-value, durable-investment, identity/personalization, collaboration/social, and comeback decisions, including exact non-applicable dispositions;
- selected monetization value exchange plus app-side refund, rewards, update adoption, cross-promotion, review/feedback, and marketing integration contracts; otherwise the exact disposition and zero-cost boundary;
- selected platform/input/HTML5/PWA, desktop OS experience/integration, offline/sync, accessibility, i18n, security/privacy, low-end quality, and experience-expression matrix, plus exact dispositions for unselected adapters;
- implementation and maintenance handoff with owners, dependency order, acceptance evidence, exposure/recovery behavior, and observed-state requirements;
- event/metric/experiment plan with selected value, quality, trust, fatigue, fairness, commercial, and support countermetrics;
- specialist handoff-request manifest with contract and acceptance tests; include digests only for already sealed inputs;
- validation roadmap, hard blockers, unresolved hypotheses, and next proofs.
Routing boundaries
app-design-blueprint owns whole-app coherence and app-side capability semantics.
saas-web-platform-blueprint owns the whole public-acquisition-to-authenticated-service composition for a SaaS platform, including tenant, commercial and operational integration contracts; a compound SaaS app may compose both with one declared primary artifact.
game-design-blueprint owns a product whose primary promise is play, mastery, fantasy, and game progression; gamification does not automatically make an app a game.
product-lifecycle-architect owns a cross-domain product program dependency and delivery graph spanning at least three lifecycle domains.
refund-and-support-flow-review owns refund consequences, repurchase, appeal, restrictions, and support; payment-platform-readiness owns provider/ledger correctness.
review-solicitation-policy owns platform-specific authentic public review
eligibility and request state; product-feedback-learning-loop owns universal
private feedback, authorized review ingestion, support/product routing, and
close-loop behavior.
- Use
notification-strategy-review, daily-reward-and-streak-review, referral-loop-review, promotion-campaign-review, product-analytics-instrumentation-review, offline-sync-conflict-review, and ad-monetization-review for their detailed artifacts.
software-distribution-readiness owns channel submission/release evidence; marketing-automation-blueprint owns the multi-channel marketing control plane.
Completion check
The blueprint is ready for composition and implementation only when another agent can identify what value the app creates, how every workflow and state behaves, why every swept capability is integrate-now, reuse-scale-ready, contract-ready, not-applicable, or floor-blocked, which selected capabilities have complete scale-ready targets, which selected legitimate features will be available without artificial gates, which conditional specialist artifacts are required, how selected HTML5/mobile/desktop and global/age modes differ, what evidence earns each proof state, and how the product degrades, recovers, and learns without leaving a future hardening or maintenance project. No selected target may be reduced or deferred solely for conventional human scarcity, and no irrelevant feature may substitute for app coherence. Blueprint acceptance does not itself claim implementation, delivery, or live availability.