agentleFS
mikezamayiasmikezamayias/agent-skills/skills/app-landing-page/SKILL.md

app-landing-page

Skill1 starsChanged 29 days agoBSD-3-ClauseView on GitHub

Generates five distinct landing page versions for an app, each a complete deployable single-file HTML. Use when creating or refreshing an app's landing page. Each version follows a different design philosophy rather than five drafts of the same page.

  • Installs packages

The file

Show its rules and text

We split this skill into the rules it gives an agent

24 rules. About 0 are common, 17 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈17

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No specific model or tool is required.

    line 31
  2. Distinctive:

    If the human asks for several models, or the environment offers other models or agents, such as subagents, another agent CLI, or a model …

    line 33
  3. Distinctive:

    Use only tools that are already installed and signed in.

    line 34
  4. Distinctive:

    Do not install or sign up for a tool without the human's approval.

    line 35
  5. Distinctive:

    Keep all 5, because the variety is the point.

    line 40
  6. Distinctive:

    Write each one as a fresh attempt from the brief and that version's philosophy, not as an edit of an earlier version.

    line 53
  7. Very specific:

    When other models or agents are available, spread the versions across them, for example one model per version, and give each one the same …

    line 55
  8. Distinctive:

    When the brief finds more than one, add these lines to it:

    line 93
  9. Distinctive:

    Do not show an iPhone frame around Android screens or the reverse.

    line 101
  10. Distinctive:

    Stop and ask the human: if you cannot confidently fill these fields.

    line 105
  11. Distinctive:

    Do not invent them. Ask for every missing field in one batch, not one question at a time.

    line 106
  12. Distinctive:

    Ask for every missing field in one batch, not one question at a time.

    line 107
  13. Very specific:

    Use base64 images only if absolutely needed, and otherwise reference images by relative path to store_assets/.

    line 123
  14. Distinctive:

    Use it as the instruction for yourself, or send it unchanged to another model.

    line 129
  15. Distinctive:

    When another model returns the page, save only the HTML and strip any commentary around it.

    line 175
  16. Distinctive:

    If no browser tool is available, open the file in a local browser and ask the human to check what you cannot measure.

    line 196
  17. Distinctive:

    Put the screenshots before the text in its message, because Anthropic's vision docs say Claude works best when images come before text, and …

    line 201
  18. Very specific:

    If its sentence does not match the version's Concept line, fix the page, not the critic.

    line 203
  19. Very specific:

    Put its scores and the sentence in the INDEX.md Critic column.

    line 205
  20. Distinctive:

    Before recording the verdict, cut the weakest decorative effect on each version.

    line 206
  21. Very specific:

    If a version fails, revise it with a targeted fix, using the same model or tool that produced it.

    line 209
  22. Very specific:

    If it still cannot be made to work, mark it Stuck in INDEX.md and move on.

    line 210
  23. Very specific:

    After all 5 are generated, write landing-pages/INDEX.md:

    line 214
  24. Distinctive:

    Write the site recipe as a skill after the first app ships, naming where assets, copy and keyword data come from.

    line 243

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

app-landing-page

This skill produces 5 distinct landing page attempts for an app. Each version is a complete deployable single-file HTML page with inline CSS and JS. The 5 versions follow 5 different design philosophies so the human can pick one, instead of 5 drafts of the same page.

When to invoke

The human says something like "make a landing page for this app", "I need 5 landing-page attempts", "draft a launch page", or "let multiple models try the landing page for X".

Activation rules

  1. The working directory must be a project root with a pubspec.yaml, project.yml, *.xcodeproj, Package.swift, settings.gradle.kts, package.json, or README.md.
  2. No specific model or tool is required. The current agent can write all 5 versions itself.
  3. If the human asks for several models, or the environment offers other models or agents, such as subagents, another agent CLI, or a model API, you may hand some versions to them for more variety. Use only tools that are already installed and signed in. Do not install or sign up for a tool without the human's approval.

The 5 design philosophies

These are 5 distinct framings with no overlap. Keep all 5, because the variety is the point.

VersionPhilosophyTypical structure
v1 - Hero-ledAbove-the-fold sells the entire pitchBig hero, 1 strong CTA, 3-feature row, footer
v2 - Story-ledLinear narrative scroll, problem → solution → proofLong-form sections, no hero CTA, and the CTA shows up mid-scroll once the case is made
v3 - Product-tour-ledAnnotated screenshots are the page, and copy supports themBig screenshot stack, captions, minimal chrome, mobile-first
v4 - Social-proof-ledTestimonials, install counts, GitHub stars front-loadedStats bar at top, quotes mid, CTA + screenshot bottom
v5 - Manifesto-ledOpinionated essay style, "why this exists", indie / founder toneWall of text, signature, single download link, no marketing language

Who writes each version

By default the current agent writes every version. Write each one as a fresh attempt from the brief and that version's philosophy, not as an edit of an earlier version.

When other models or agents are available, spread the versions across them, for example one model per version, and give each one the same prompt template. Record which model or tool produced each version in INDEX.md.

Project discovery (Phase 0)

Build the brief from these files only:

  • pubspec.yaml → name, description (Flutter)
  • project.yml → name, bundle ID prefix, deployment targets, Info.plist display name (native Apple)
  • app/build.gradle.kts → applicationId, minSdk, and res/values/strings.xml app_name (native Android)
  • A distribution file such as a Homebrew cask or release script → direct download URL and Mac requirements
  • README.md → first heading + first paragraph
  • store_assets/BRAND_PROFILE.md (if app-graphics ran already) → colors, voice
  • store_assets/graphics/app_icon/app_store_1024x1024.png (if exists) → reference for visual style
  • store_assets/screenshots/framed/en-US/ (if exists) → screenshots to embed

Write landing-pages/LANDING_BRIEF.md:

# <App> landing brief (built <YYYY-MM-DD>)

- Name: …
- One-liner: …
- Target audience: …
- Top three objections that keep a visitor from converting: …
- Traffic source: <store listing, ads, search, social, email, direct>
- Conversion event: <what counts, for example App Store tap, TestFlight join, email signup>
- Primary CTA: <e.g. "Join the TestFlight", "Download on the App Store", "Get on GitHub">
- Anchor colors: …
- Voice: …
- Key proof points (max 5): …
- Available screenshots: <relative paths or "none">

Multi-platform native apps

Flutter apps usually have one CTA pair. A native project can ship several products from one repository, such as an iPhone app with a Watch app and widgets, a Mac app outside the App Store and an Android app. When the brief finds more than one, add these lines to it:

  • Platforms and download path for each: App Store, Google Play, TestFlight, or a direct Mac download (notarized, with a Homebrew cask command such as brew install --cask <name>).
  • Which platform the primary CTA serves. Detect the visitor's platform with navigator.userAgent or CSS-free link logic, show that platform's button first, and keep the others visible as secondary links. The page must still work when detection fails.
  • Mac requirements (minimum macOS, Apple silicon or Intel), shown near the Mac download button.
  • Screenshots per platform from app-screenshots, with the Watch and Mac shots labelled by device. Do not show an iPhone frame around Android screens or the reverse.
  • SoftwareApplication JSON-LD per platform with its own operatingSystem and real URL. See references/seo-and-ai-answers.md.

Stop and ask the human if you cannot confidently fill these fields. Do not invent them. Ask for every missing field in one batch, not one question at a time.

Output structure

landing-pages/
├── LANDING_BRIEF.md
├── INDEX.md                  # which model or tool produced which version + verdict
├── v1-hero-led.html
├── v2-story-led.html
├── v3-product-tour-led.html
├── v4-social-proof-led.html
└── v5-manifesto-led.html

Each .html is a single file with inline CSS and inline JS. Use base64 images only if absolutely needed, and otherwise reference images by relative path to store_assets/. There is no build step, so the file works when dropped on any static host, such as Cloudflare Pages, Vercel, or GitHub Pages.

Per-version prompt template

Build the prompt for each version from the brief and the version's design philosophy. Use it as the instruction for yourself, or send it unchanged to another model.

Build a single-file HTML+CSS+JS landing page for "<App name>".

One-liner: <…>
Target audience: <…>
Primary CTA: <…>
Anchor palette: <hex codes>
Voice: <…>

Design philosophy for THIS version: <one of the 5 above, copied from the table>

Constraints:
- Single .html file. Inline all CSS and JS. No external dependencies except Google Fonts (one family max) and an optional inline SVG icon set.
- Mobile-first responsive (works at 375×667).
- One primary CTA on the page; secondary CTAs allowed but visually subordinate.
- If the page receives TikTok or Instagram bio traffic, the primary CTA must still work inside their in-app browser. Add a full-screen overlay with an arrow to the ••• menu and "Tap ••• → Open in browser", and keep a visible Download button. One founder's claim (2026-10-20), untested, see the link-in-bio notes in `ship-first-dollar-app` references/distribution-channels.md.
- A page that gets social link traffic can be mostly phone.
  One developer-tool founder reports 82 percent mobile, 17 percent desktop and 1 percent tablet visitors after an X link post, and attributes a rise in bounce rate to it (single report, the attribution is theirs and untested).
  Check the primary CTA and the first screen at phone width, and read the mobile share for each traffic source in the analytics after launch.
- Real copy only. No "Lorem ipsum", no placeholders.
- No marketing puffery: no "transformative", "revolutionary", "the best", "simplify your life", "unlock the power of". Write headlines in plain, specific language.
- No AI-copy tells: no "where X meets Y", no "not just X but Y", no "scroll to explore", and no "hover" in copy a touch visitor reads.
- Avoid unless the concept demands it, and then say why in INDEX.md: a centered hero over a gradient blob, purple-blue gradients, cream with terracotta serif editorial, acid green on black, glassmorphism, an eyebrow label over every heading, and 01/02/03 numbering on content that is not a sequence.
- Light mode default; dark mode via `prefers-color-scheme: dark` if the philosophy fits.
- Screenshots: <relative paths from LANDING_BRIEF, or "none, design around it">
- Accessibility: semantic landmarks, alt text, color contrast WCAG AA.
- Concept: one sentence naming what this page is about that only this app could own, anchored in a real fact from the brief. If the sentence still works with another app's name swapped in, it is a style, not a concept. Carry it through every section, not just the hero.
- Type: body 16-18px with 1.45-1.6 line height, 45-70 characters per line, tabular figures for numbers, a non-breaking space between a number and its unit.
- Motion: animate only transform, opacity and clip-path. Under `prefers-reduced-motion`, all content must be visible with no scroll-driven animation.
- Autoplaying loops: motion that starts by itself, lasts more than 5 seconds and runs beside other content needs a control to pause, stop or hide it (WCAG 2.2 criterion 2.2.2, Level A), unless the motion is essential. Gating it behind `prefers-reduced-motion` does not satisfy 2.2.2. Content that updates automatically has no 5 second exception, and a control for the update frequency also satisfies it.
- Interaction: primary targets (CTA, nav, screenshot links) answer with one movement treatment per role (label roll, underline draw, fill, arrow nudge), never only a color change. Legal and social links may only change color. Put every :hover inside @media (hover: hover), give the same state to :focus-visible, and under prefers-reduced-motion keep a visible color or underline state instead of the movement.
- No em or en dashes in copy. Exactly one <h1>.
- The top third says what the app is, who it is for and what to do next. The last CTA on the page uses the same label as the first. A logo row holds at most 6 logos. Every metric carries a visible source line.

Return only the complete HTML file. No commentary.

The last constraint line is adapted from the taste-landing-page skill in madebymustafa/design-taste (MIT, read at commit c4a6bff, 2026-10-16). Those are that skill's own rules, not findings from conversion research, so treat them as defaults and drop one when a version's philosophy needs to.

The copy tells, the avoid list and the movement-per-role rule in the template are adapted from juliobsz/skills /awwwards (MIT, read at commit 461fab3, 2026-10-16), where the audit script flags the copy tells and the skill lists the avoid list. The hover gating and reduced-motion lines rest on the CSS Media Queries Level 5 hover feature and MDN prefers-reduced-motion, not on that skill. hover: hover means the primary pointing device can easily hover, and the query tests only the primary input.

When another model returns the page, save only the HTML and strip any commentary around it.

Verification per version

For each generated HTML, measure in a real browser (agent-browser, the T3 preview, or Chrome DevTools MCP as in web-perf) rather than eyeballing:

  • Zero console errors and zero failed requests.
  • No horizontal overflow at 375×667. Name the offending element if there is one.
  • Zero serious or critical axe-core violations at WCAG 2.1 AA, which also measures contrast ratios.
  • Every tap target at least 24×24px.
  • Keyboard: every focus stop visible, tab order follows the page, the CTA reachable.
  • Reduced motion: emulate prefers-reduced-motion: reduce and confirm no text stays hidden.
  • Throttled mobile (4× CPU, slow 4G): Largest Contentful Paint under 2.5s and Cumulative Layout Shift under 0.05.
  • Text audit: no "Lorem ipsum", TODO outside the CTA, or bracketed placeholders; no repeated sentences; none of the copy tells "where X meets Y", "not just X but Y" or "scroll to explore"; exactly one <h1>; every image has alt and every link a name.
  • Hover: capture each primary target at rest and under the pointer at the same clip. Fail any primary target with no visible change, and read every color-only change by eye. A full fill or inversion passes, and a text-color swap on a CTA or nav link does not.
  • Idle: leave the pointer still for five seconds on desktop. If nothing that carries the concept moves, write "still by decision" in INDEX.md, or add a small idle motion on the concept itself, never a decorative floating shape. Any idle loop that runs more than 5 seconds beside other content follows the autoplaying loops rule above.
  • Primary CTA is wired to a real URL, or to # TODO: … clearly marked if not yet known.
  • No external script tags from untrusted CDNs (audit <script src=).
  • Search and AI-answer ship checks from references/seo-and-ai-answers.md pass for the chosen version, which also holds the recurring SEO loop.

Record any check you could not run as not verified in INDEX.md, rather than passing it. If no browser tool is available, open the file in a local browser and ask the human to check what you cannot measure.

Then run a fresh critic: a new subagent that sees only the version's mobile and desktop screenshots and the check results, not the prompt or this conversation. It writes one sentence on what the page is about, then scores Design, Usability, Creativity and Content from 1 to 10, starting each at 6 and citing a screenshot or check result for every point above it, and lists the top 3 fixes. Give it the rest and hover screenshot pairs from the Hover check too. Put the screenshots before the text in its message, because Anthropic's vision docs say Claude works best when images come before text, and describe image-then-text as a preference since other orders still perform well. Launch a new critic after every revision, so it never anchors on an earlier score. If its sentence does not match the version's Concept line, fix the page, not the critic. Dispute a score only by naming the screenshot it misread. Put its scores and the sentence in the INDEX.md Critic column. The human still picks the winner. Before recording the verdict, cut the weakest decorative effect on each version. Hover, press and focus feedback are not decoration and stay. Thresholds and the critic method are adapted from juliobsz/skills /awwwards (MIT, read at commit 461fab3, 2026-10-16).

If a version fails, revise it with a targeted fix, using the same model or tool that produced it. If it still cannot be made to work, mark it Stuck in INDEX.md and move on.

Write the INDEX

After all 5 are generated, write landing-pages/INDEX.md:

# Landing page versions for <App name> (built <YYYY-MM-DD>)

| Version | Philosophy       | Model or tool | Words | Critic (scores + sentence) | Verdict (human fill) | Path                                   |
| ------- | ---------------- | ------------- | ----- | -------------------------- | -------------------- | -------------------------------------- |
| v1      | Hero-led         | <model>       | XXX   |                            |                      | landing-pages/v1-hero-led.html         |
| v2      | Story-led        | <model>       | XXX   |                            |                      | landing-pages/v2-story-led.html        |
| v3      | Product-tour-led | <model>       | XXX   |                            |                      | landing-pages/v3-product-tour-led.html |
| v4      | Social-proof-led | <model>       | XXX   |                            |                      | landing-pages/v4-social-proof-led.html |
| v5      | Manifesto-led    | <model>       | XXX   |                            |                      | landing-pages/v5-manifesto-led.html    |

## Stuck / skipped

- (list any version that did not generate, with the model and the failure reason)

Run log

Append one line to landing-pages/runs-log.md:

YYYY-MM-DD HH:MM - <project> - 5 versions (<model> x N, ...)

Several apps

For several apps use one Astro codebase with one content collection per app, so shared changes (crawler rules, JSON-LD, localization) land once. Write the site recipe as a skill after the first app ships, naming where assets, copy and keyword data come from. This is a preference stated in a video by Dan Kulkov (2026-10-19), not a comparison against other frameworks, and it applies to a site that outgrows the single-file pages above.

Related skills

  • app-graphics generates the icon and OG card this landing page references.
  • app-screenshots generates the framed screenshots this landing page embeds.
  • For a "how it works" figure, use the anatomy skill, at most one per page and only when the page must explain a mechanism, with every fact taken from the brief and never invented.
  • For an "AI is building this" reveal on a feature card, read references/ai-generation-effect.md first. Never on the hero, unless a throttled lab run with the effect on the hero element keeps Largest Contentful Paint at or under 2.5 s.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.