agentleFS
MGeorge98MGeorge98/claude-skills/landing-studio/SKILL.md

landing-studio

Skill0 starsChanged 39 days agoMITView on GitHub

Orchestrator playbook for a story-driven landing page that a brand team would ship: a Next.js 15 app built from this skill's kit (attribute-driven motion engine, pinned scroll scenes, word masks, brand cascades), with a lead API and a branded confirmation email, every claim traced to a facts ledger, and verified by audit scripts (fold, scrollWidth, flick, paints, CLS, LCP, JS ceiling, facts gate) plus contact sheets you look at. Use when the user says "fă-mi un landing", "landing page cu poveste", "pagină de prezentare", "landing breathtaking", "scroll storytelling", "pagina din QR", "landing for this event / deck / product", "a page people land on from the QR code", or asks for a scroll-animated one-pager with a form.

  • Reads credentials
  • Installs packages
  • Commits and pushes

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Not for multi-page sites, dashboards, or cold-outreach pitch packages (that is the pitch-app pipeline in this repo, with its own gate).

    line 37
  2. Very specific:

    Finished page without motion: No JS, reduced motion, and a dead bundle all render the settled composition (html.js:not(.rm) gates reveals, …

    line 49
  3. Distinctive:

    Nothing outward by an agent: no deploy, no email to a real person, no push, no env on Vercel.

    line 52
  4. Very specific:

    When a source changes (new deck, edited bio), do not patch copy by memory.

    line 94
  5. Very specific:

    Run against a production build: npm run build && PORT=3050 npm start < /dev/null &.

    line 143
  6. Distinctive:

    Every automated check here has passed a blank section at least once.

    line 159
  7. Distinctive:

    After Phase 0: the brief, ten lines.

    line 196

≈ “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

Landing Studio

Goal: one page, one action, told as a story in scenes, with zero visual bugs and zero invented facts. The method is: brief, facts, brand, storyboard, one reference scene, fan out, audit from pixels, ship through the owner. Reference build: the MBV Summit QR page (~/Downloads/SPEECH, 2026-10-08).

Language: this playbook is English; the page's language comes from the request or the source deck ("fă-mi..." means Romanian copy with full diacritics). Set <html lang>.

Files in this skill:

  • CONTRACT.md: the kit contract (boot classes, [data-rv], [data-scene] and --p, tokens, prefixes). Agents read it.
  • kit/engine/, kit/css/, kit/react/: the motion engine and its CSS (kit/README.md); kit/scenes/<name>/ with kit/scenes/INDEX.md (the scene catalogue); kit/backend/ (lead API, email, .env.example); kit/app-template/ (Next.js 15 shell with the boot script) and its sync-kit.sh.
  • scripts/audit.cjs, scripts/perf.cjs, scripts/gate.cjs: the audit (scripts/README.md, references/audit-targets.md).
  • gallery/README.md: reference builds with their contact sheets.
  • references/: brief-template.md, facts-ledger.md, storyboard-template.md, motion-language.md, parts-contract.md, audit-targets.md, gotchas.md, cliches.md, handoff-template.md.

When to use

  • A page whose job is one action (leave contact, book, download, join) and whose material is a story: an event, a talk, a product launch, a founder, a pitch deck turned web page.
  • Not for multi-page sites, dashboards, or cold-outreach pitch packages (that is the pitch-app pipeline in this repo, with its own gate).

Fixed decisions (write them into brief.md; agents obey them)

  • Every fact traces to a source line. facts.md is the only place copy may take a claim from. INFERRED lines never ship. No invented numbers, testimonials, logos, ratings, dates or deadlines.
  • No em dashes anywhere: copy, alt text, email, subject, code comments in shipped files.
  • One action. The form (or the button) is the page's destination; every scene points at it.
  • One orchestrated moment beats scattered fades. Each scene has one idea and one motion.
  • The kit is the engine. Scenes use CONTRACT.md attributes; no new scroll listeners, no GSAP, no Lenis, no getBoundingClientRect in a frame. Transforms, opacity and clip-path only.
  • Finished page without motion. No JS, reduced motion, and a dead bundle all render the settled composition (html.js:not(.rm) gates reveals, html.mo:not(.rm) gates pins; 3.5 s failsafe).
  • Pages that carry someone else's brand ship noindex, nofollow until the owner says otherwise.
  • Nothing outward by an agent: no deploy, no email to a real person, no push, no env on Vercel.

Phases

Model routing: Opus for the facts ledger, storyboard directions, the reference scene, the integrator and the final LOOK; Sonnet for brand ingest, scaffold, scene fan-out, audit runs and fixes with a clear finding. Never rerun a failed brief on the same model unchanged: fix the brief or escalate.

PhaseWhoWhatDone means
0 Briefyoubrief.md from the template below: 10 lines, the three dials, the one action, the sourcesthe owner has read it (✋ only if a line is a guess)
1 Facts1 agent, Opusfacts.md per references/facts-ledger.md: every claim the page could use, one line each, with file:line or URLevery line has a source and a status; INFERRED lines flagged
2 Brand1 agent, Sonnettokens.css (--bg --ink --ink-2 --muted --accent --accent-2 --line, fonts), logos as SVG plus dark-pill PNGs for email, font files localtokens contrast-checked (body 7:1, small 4.5:1); logo renders on light, dark and inverted
3 Storyboard1 agent, Opusthree directions per references/storyboard-template.md, differing in metaphor, not palette; scene order from kit/scenes/INDEX.md; copy slots filled with fact ids; self-review against references/cliches.mdthree tables plus one still per direction
✋ownerpicks a direction (and edits copy)choice written into brief.md
4 Scaffold1 agent, Sonnetcopy the template, sync-kit.sh with the scenes and --backend (commands below), brand values into src/styles/tokens.css, facts into src/config.tsnpm run build green, /preview/<scene> renders at 390 and 1440, form posts to data/leads.jsonl
5 BuildOpus (reference) then 3-4 Sonnet agentsthe reference scene first, then one scene folder per agent (clones, ports, prefixes below); the integrator alone writes src/app/page.tsxeach scene passes its acceptance list in references/parts-contract.md; integrated page builds
6 Audit1-2 agents, Sonnet; LOOK by you or Opusaudit.cjs, perf.cjs, gate.cjs --facts facts.md, email preview, contact sheets; regenerate only failing scenesall targets in references/audit-targets.md green; every sheet looked at; findings table closed
✋ownerreview the live local page and the email previewapproved or a change list
7 Shipyoucommit per wave (pathspec), handoff note, owner-only listHANDOFF.md written; commands handed over, none run

Phase 0: the brief (copy into <app>/brief.md)

1  Page:        <name>, <one line: who lands here, from where (QR, link, ad)>
2  Audience:    <who, what they already know when they land>
3  One action:  <the single thing they do: leave name+email+phone / book / download>
4  Promise:     <what they get for it, in their words>
5  Story:       <the arc in one sentence: from X to Y>
6  Tone:        <three words>, voice <tu / dumneavoastră / you>
7  Dials:       DESIGN_VARIANCE <1-10> · MOTION_INTENSITY <1-10> · VISUAL_DENSITY <1-10>
8  Sources:     <deck path, site URL, bio doc, brand files> (facts come only from these)
9  Constraints: <deadline, hosting, language, noindex, what must not appear>
10 Done:        <the owner's own test, e.g. "scan the QR on a phone, submit, get the email">

Dials: VARIANCE 3 = familiar grid, 8 = a layout nobody has seen; MOTION 3 = reveals only, 8 = pinned scenes that carry the story; DENSITY 3 = one idea per screen, 8 = a dense editorial page. Write the numbers before the storyboard; agents read them as constraints. Details: references/brief-template.md.

Phase 1: facts and reconciliation

When a source changes (new deck, edited bio), do not patch copy by memory. Diff old source against new, then produce a table file:line in the app | current text | source line | action and apply it. Re-run gate.cjs --facts facts.md after. The SPEECH build needed this twice in one day.

Phase 3: storyboard self-review (before any code)

Each direction is a table (order | scene | copy slots to fact ids | mobile note | frames to verify). The storyboard agent checks every direction against references/cliches.md and rewrites hits before you see it. The hero frame states the hook above the fold at 390x844. If the form is the action and the audience arrives warm (QR at an event), put the form in the first screen.

Phase 4: scaffold commands

SK=~/.claude/skills/landing-studio; APP=<repo>/<slug>          # brief.md, facts.md already in $APP
mkdir -p "$APP" && cp -R "$SK/kit/app-template/." "$APP/" && rm "$APP/sync-kit.sh"
"$SK/kit/app-template/sync-kit.sh" "$APP" --scenes hero-words,pinned-beats,form-popup --backend   # any of the 16 names in kit/scenes/INDEX.md
cd "$APP" && npm install && xattr -dr com.apple.quarantine node_modules 2>/dev/null; npm run build
  • sync-kit.sh lays out src/kit/{engine,react}, src/styles/{tokens,motion}.css, src/scenes/<name>/, src/app/preview/<name>/page.tsx (the scene alone between two 100svh spacers, with its data.example.ts) and, with --backend, src/app/api/lead/route.ts, src/lib/{leads,email}.ts, src/config.ts, .env.example. It is safe to re-run after a kit fix: kit code is overwritten, app-owned files (tokens.css, config.ts, previews) are written only when missing. Always run it from the skill, never from a copy.
  • node_modules: when another local app has the same package.json dependency set, cp -Rc <that>/node_modules . (APFS clone) instead of npm install, then npm ls --depth=0 to confirm.
  • Brand (Phase 2 values): edit the brand block of src/styles/tokens.css (--bg ... --font-mono) and THEME_COLOR in src/app/layout.tsx (same as --bg). Facts for the backend go into src/config.ts.
  • Already in the template: robots: { index: false, follow: false } and an X-Robots-Tag: noindex, nofollow header in next.config.ts; remove both only when the owner says the page may be indexed.
  • Assets the scenes reference (public/<shot>.png, public/email-logo-<id>.png) are copied by hand.

Phase 5: build

  • Reference scene first, Opus: the hardest scene in the storyboard (usually the hero or the first pinned scene), built to the full acceptance list. Its file becomes the example in every later brief.
  • Fan-out: one agent per scene folder src/scenes/<name>/ (new scenes go back into the skill as kit/scenes/<name>/ with scene.md and preview.tsx), classes prefixed .<name>-, nothing else touched. Each agent works in its own clone: cp -Rc "$APP" "$TMP/<name>" (APFS clone, node_modules included; never a symlink, Turbopack refuses it) and runs on its own port: PORT=3051 npm run dev < /dev/null (3052, 3053...; next dev exits when stdin closes). The integrator runs on 3050. Each scene is checked alone at /preview/<name>.
  • Integrator (you or one Opus agent) copies finished scene folders back, writes src/app/page.tsx, resolves seams between scenes (background handoffs, first frame of each pin visible at p=0).
  • A kit bug is fixed in the skill (kit/...) and re-synced with sync-kit.sh, never patched in the app only.
  • Rules: references/parts-contract.md. Motion: references/motion-language.md.

Phase 6: audit

Run against a production build: npm run build && PORT=3050 npm start < /dev/null &. Kill an old server by port first (lsof -ti tcp:3050 | xargs kill): a stale server answers with the old build and every script measures it. Write ls-audit.json in the app (format in scripts/README.md: url, each pinned scene's selector and frames, hero fields that must sit in the first fold). Then, from the app folder:

SK=~/.claude/skills/landing-studio; K=$SK/scripts
node $K/audit.cjs                  # profiles m 390, d 1440, rm, js-blocked; --scene <sel|i> to rerun one scene
node $K/perf.cjs                   # paints per scene <= 120, idle <= 3, CLS <= .02, LCP, JS vs perf-baseline.json
node $K/gate.cjs src --facts facts.md   # em dash / placeholders fail; unsourced names and numbers warn (--strict fails)
node --experimental-strip-types $SK/kit/backend/preview/render-email.mjs   # preview-out/: 640 and 390, light, dark, invert

Playwright comes from PW_PATH / NODE_PATH (scripts/README.md). The first perf.cjs run writes perf-baseline.json; commit it, never raise it. Look at audit/sheet.png and every scene frame plus the email shots. audit.cjs frames pinned scenes only; a flow scene (reveal-driven) is checked from hand-shot frames (scripts/shoot.cjs <url> --full, or a playwright script sampling 0 to 1400 ms after it enters the viewport). Every automated check here has passed a blank section at least once. Failing scenes go back to their own agent with the finding; passing scenes are frozen. Last gate: the category audit (would this page stand next to the three best pages in its category; name the gap or say none). Before ship: rm -r src/app/preview (and the data.example.ts files the page does not import).

Phase 7: ship

Main session alone touches git: git add <app>/src/scenes/<name> ... per wave, never -A. Handoff from references/handoff-template.md. Then hand over, never run:

Owner-only (hand over the exact commands)

  • Env vars on Vercel (SMTP, notify address, any key) and vercel env pull (it can carry a real DB URL).
  • vercel deploy / --prod, domains, DNS. Auto mode denies deploy; do not work around it.
  • The real submit test on a phone from the real URL, and the first real email to a real person.
  • git push, unless the owner asked for it in this session. Deleting worktrees and clones.

Sub-agent brief skeleton

Goal: <one sentence + why it matters to the page>
Scope: you own src/scenes/<name>/ and src/app/preview/<name>/ in your clone <path>, port <30xx>.
  Off limits: src/kit/, src/styles/, src/app/page.tsx, facts.md, every other scene. Never git, deploy or send.
Known facts: CONTRACT.md; storyboard row <n>; copy slots -> F-ids <list> (quote facts.md exactly);
  dials <V/M/D>; reference scene <path>; gotchas that bite this scene: <3 lines from gotchas.md>.
Done means: acceptance list in references/parts-contract.md, all items, at 390x844 and 1440x900;
  you looked at your sheets; server stopped.
Report in under 150 words: files, acceptance results (n/n), sheet paths, open issues. No code.

Done means

A brand team would ship it; the audit is green; you looked at every sheet. Short of that, it is not done, whatever the reports say. "Done" in an agent report is a claim: re-run audit.cjs yourself.

Talking to the owner

  1. After Phase 0: the brief, ten lines. Proceed unless they object.
  2. ✋ After Phase 3: three directions, one still and one paragraph each, your recommendation.
  3. ✋ After Phase 6: local URL, the sheets, the email preview, the audit summary.
  4. At the end: what is committed, the owner-only commands, what is still open.

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.