More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Concept Design: Distinct Mobile App Screens, Not Templates

    1

    Turns a product brief into three distinct mobile app design directions, a full screen suite and a written contrast and tap-target checklist.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Concept Design: Distinct Mobile App Screens, Not Templates

    Concept Design: Distinct Mobile App Screens, Not Templates

    Example session with this skill installed

    Design the home screen and one detail screen for my iPhone app "Fernlog", a plant-care companion for people with 5 to 20 houseplants. They want to see who needs watering today and whether each plant is recovering. Use /concept-design. Show me three different directions first and recommend one. Everything in the app can be invented for now.

    • Read your context and instructions
    • Compiled the concept design

    Fernlog: home + plant detail. All names, plants and data are invented for this example.

    Concepts. I read the brief for what people act on and worry about. Seven raw ideas, merged to three: time-to-care (who needs me today), place (rooms and their light), growth (is it recovering). A herbarium-card idea was rejected because it was an invented paper metaphor that hid the thing people act on.

    Three directions, same home screen, same content, same scale

    A "Sill". Plants grouped by the room they live in, each standing on a drawn sill ledge, with a daylight card. Weakness: "what do I do now?" is answered by chips, not a list.

    B "Level". Sorted by when each plant needs you. Every plant's moisture is the fill height inside a pot silhouette, repeated in the hero, rows and detail. The calmest layout.

    C "Journal". A dark visual diary of portrait cards, each with a one-line change ("+2 leaves this month") and a 12-week health trace. Weakness: today's tasks are buried.

    Recommendation: B "Level". It answers "who needs me?" in the first 400 px, its pot gauge belongs to the subject and survives with the hero removed, it scales to 20 plants, and its water-teal accent breaks the category's green-on-green sameness. This is my choice as the author, not your approval; A and C stay available, and I took A's light reading and C's health trace into the detail screen.

    Colour pairs, direction B (WCAG, computed with the checklist formula)

    RoleForegroundGroundRatioNeed
    Row title on card#17211C#FFFFFF16.53:14.5
    Secondary on card#566059#FFFFFF6.54:14.5
    Water status text on card#0E6B76#FFFFFF6.21:14.5
    Active tab on capsule#0E6B76#E8F1F25.41:14.5
    Task ring (control)#78837C#FFFFFF3.94:13.0

    The brand teal #1E8592 is 4.35:1 on white, so it fills the gauge but never sets text.

    Delivered. 01-home-today, 02-plant-maple, 03-home-first-use (HTML and 1206 x 2622 PNGs), a shared kit stylesheet, a comparison board, states.md, review.md.

    Checklist, final round, 6 pages. 0 FAIL. Round one found and fixed: a task ring at 2.61:1 (now 3.94:1), text under a scroll fade at 3.01:1 (fade shortened, now 6.54:1), a 13 px banner line (now 15 px), 46 off-grid spacing values in direction C (now 0). Standing: 3 accepted WARN (two exploration frames with a fade over the last row; a two-colour gauge at 3.55:1 that the checklist does not cover), 2 checks not run (the bottom safe area, not measured; kit stylesheet on the exploration frames, by design).

    Not done, and why. No simulator or device run, no VoiceOver, no dark mode or loading and error states drawn. The reference study used marketing pages because no real captures of the leaders were reachable, so every number in study.md is labelled estimated.

    Next steps. 1) Pick A, B, C or a merge. 2) Add loading and error states for home. 3) Review the screens on a real device before they replace anything. 4) Ask for application code as a separate request.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    AI-made app screens all look like the same template

    Ask an agent for "a clean fintech home screen" and you get the same purple gradient card, the same four rounded icon buttons, the same list of transactions. It is usable. It could belong to any product, and nobody remembers it.

    Concept Design makes your agent design the way a studio does: from the product outward, with a study of how leading apps are built, several genuinely different directions to choose from, and a checklist that catches the small failures (grey text that is too faint, tap targets too small, a status bar typed as text) before you see the work.

    What it does, step by step

    1. Brief. Audience, the one action that matters, language and direction, platform and content, written down. For an existing product it first maps every screen, sheet and state.
    2. Reference study. Before drawing, it studies leading shipped apps for how they are built (bar heights, margins, radii, type sizes) and compares your frame with theirs at the same scale, naming measured differences. It never copies another app's identity, assets or text.
    3. Concept. Ideas come from your product's own content, people and tensions, not from a stock metaphor. Similar ideas are merged, not padded into a count.
    4. Three directions. The same screen, same content, same size, in each direction, with a card (idea, type, colour roles, trade-offs) and a table of every text and icon colour against its ground with its contrast ratio. When redesigning, today's screen sits beside them. You choose or merge; a recommendation is never treated as consent.
    5. Shared kit, then the suite. One stylesheet of tokens and parts that every screen links, then every screen and relevant state (empty, loading, error, offline), with a state list of trigger, action and destination.
    6. Critique. Two separate checks, usability and visual authorship, with every PNG opened at full size and the findings written down.
    7. Checklist. 27 written checks applied to every screen and direction (contrast, tap targets, text sizes, drift in radii, sizes and spacing, RTL, status bar, safe area, whether directions differ on structure and more), each with a threshold, worst case reported.
    8. Delivery. Editable HTML, phone-sized PNGs, a comparison, the review and the limits, next to your old look until you have seen the new one on a real device.

    What you get

    • A workflow your agent follows from SKILL.md, with 13 reference files: 71 design rules and 82 finish rules, each with an id so reviews can cite them.
    • A written checklist with the exact thresholds and a console snippet that lists every text element's contrast, size and tap box on a page.
    • Templates: a device-chrome stylesheet (status bar, home indicator drawn from measured values), a direction page shell and an exploration record.
    • A licence that lets you use it for your own and your clients' projects. What you design with it is yours.

    Who it is for

    • Developers building with agents who want a real visual direction instead of the default look, without hiring a designer first.
    • Designers and small studios who want a fast first round of distinct directions and a systematic review pass.
    • Founders who need to see a few directions for their product before committing to one.

    What it does not do

    • It does not write application code. It produces HTML design studies and PNGs; turning them into your app is a separate request.
    • It does not copy other apps. The reference study measures how they are built; the output is drawn from your product.
    • It does not replace a native-device test, VoiceOver, Dynamic Type or an accessibility audit.

    Honest limits

    • Text-only edition. There are no scripts and no images in this package. The checklist is applied by your agent reading the page source and the PNGs, using a console snippet in a browser, so its verdicts are as careful as that reading.
    • Captures need a browser. Your agent has to render each page to a PNG at a true 402 x 874 viewport; the package explains how, but it needs a browser or a capture feature in your agent. Without one you get the HTML and text guidance, and the skill marks visual verification as incomplete.
    • The reference study needs real captures of leading apps. If an agent can only reach marketing pages, the skill requires it to say so and to label every number as estimated.
    • Measured on an iPhone-sized viewport. Android system chrome is not specified.
    • Tested with Claude Code on macOS. Other agents are untested. It has been run end to end once, blind, by an agent that had never seen it, on a brief for an invented plant-care app (the sample output below). Wider testing is not done.
    • A clean checklist is a measured floor, not proof of a good design. Taste review stays with you.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Mobile Development.

    • No scripts or images ship: the checklist is applied by the agent reading page source and PNGs, with a console snippet; verdicts are as careful as that reading.
    • PNG capture and the console snippet need a browser the agent can drive.
    • The reference study needs real captures of leading apps; when only marketing pages are reachable, numbers are labelled estimated and the skill says so.
    • Measured on an iPhone-sized viewport (402 x 874); Android system chrome is not specified. The bottom safe area is not measured and is reported as not checked.
    • Browser simulation, not a native-device test; no VoiceOver or Dynamic Type check.
    • Some rules pull against each other on certain screens (for example a type scale against density); the skill has the agent record the trade-off.
    • Tested with Claude Code on macOS, run end to end once; other agents untested.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 20 days ago

    Nothing to install. It runs in your AI agent.

    Optional
    Web browserInternet access

    The browser is for PNG captures and the console snippet; Chromium-based is recommended. Internet helps the reference study of public app pages.

    Permissions

    Read Files
    Write Files
    Network Access
    Browser

    File Scopes

    design/**

    Reads existing screens and tokens in your project to map them (read only). Writes only the study folder design/<date>-<subject>/. Browser is optional, for PNG captures. Network use is limited to the reference study, which reads public pages of leading apps; no terminal or shell, no environment variables, no services, no account or API keys.

    Listed23 days ago
    Updated20 days ago

    What's inside

    Frequently Asked Questions

    No. It studies how leading apps are built (heights, margins, radii, type sizes) so your design reaches the same level of finish, then names the differences. Logos, brand colours, illustrations, copy, icons and a leader's signature mechanism are never copied, and their screenshots are never shipped in your work.

    Yes, as a method. For Arabic interfaces it renders right-to-left first, keeps existing Arabic terminology, requires dir and lang on the page, and the checklist fails a screen that is mostly Arabic and not marked right-to-left (dir=rtl). The worked Arabic screens in the screenshots have not yet been reviewed by a native speaker; have your own wording reviewed.

    No. The output is editable HTML and PNG renders that your agent writes into your project. Nothing connects to Figma, and it does not export to it.

    27 checks at a 402 x 874 viewport: text and icon contrast on the painted ground (4.5:1 and 3:1), tap targets of 44 x 44 px, running text of 15 px, drift in radii, font sizes and spacing, a shared kit, placeholder text, RTL, status bar and safe area (for the device's own insets), icons drawn rather than typed, type families, capitals, monospace misuse, whether two captures of a screen are identical, and whether the directions differ on structure and show their concept.

    Yes. For an existing product it first maps every route, screen, sheet and state from your code or the running app, bounds the run by that map, and shows today's screen as frame 00 beside the new directions. It delivers the new look next to the old one, which it leaves unchanged. It does not edit your application code.

    No. It delivers HTML study screens, PNGs, a comparison and a review. Implementing the chosen direction in your framework is a separate request to your agent.

    You get the written method and every reference, but the agent cannot render PNGs or run the console snippet, so the checklist cannot be fully applied. The skill then hands over the HTML and states that visual verification is incomplete.

    Tested with Claude Code on macOS only. It is plain markdown, so other agents that load skills may follow it, but that is untested.

    New in Mobile Development