PageSmith Studio — Pages That Look Human, Convert, Get Cited

Your agent can build a landing page in a minute. Visitors can tell in two seconds.

Soft blue, purple gradient, three feature cards, "Welcome to the future of…", a testimonial from "John D., CEO". Every coding agent ships the same page by default, and the same page gets the same result: visitors bounce, and AI search engines never cite it.

See it before you buy

The skill's own site was built with the skill: https://pagesmith-studio.ercole-devs.workers.dev/

On that page you can drag a before/after slider between the default page an agent ships for a brief and the page PageSmith Studio ships for the same brief, and open three complete demo sites built with the method, each with its own photography, typography, structured data and a 3D scene:

• a wood-fired bakery in Trento (Italian, local business): https://pagesmith-studio.ercole-devs.workers.dev/demos/forno-bianchi/
• rota software for restaurants (English, SaaS): https://pagesmith-studio.ercole-devs.workers.dev/demos/turno/
• a physiotherapy clinic in Bologna (Italian, service): https://pagesmith-studio.ercole-devs.workers.dev/demos/fisio-ponte/

The "before" page scores 0/100 on the skill's tell check. The three demos score 100/100. Same brief, same model: the only difference is the order of operations.

How it works

Give it a URL or answer eight questions. Your agent runs:

1. Brief — extracts the offer, audience, proof and stack from your site; asks only for the gaps
2. Research — category leaders, the colours that physically exist in your business, your customers' own words, the ten questions people ask ChatGPT and Perplexity about businesses like yours (three tracks in parallel when your agent supports subagents)
3. Three design directions — scored, one winner, written rationale, full design system
4. Conversion copy — message hierarchy, three hero candidates, every section, microcopy, CTA strategy, a "claims to confirm" list; real proof only, never an invented number or testimonial
5. Build — static HTML/CSS, Astro, or your existing Next.js/Nuxt/SvelteKit; full content in the HTML because AI crawlers do not run JavaScript; accessible by default
6. SEO + GEO layer — metadata, JSON-LD generated from your brief, answer blocks that AI engines can lift verbatim, llms.txt, a robots policy that lets AI crawlers in, and a test that your CDN is not returning 403 to GPTBot, ClaudeBot and PerplexityBot
7. Five quality gates — Tells, CRO, GEO, Performance & accessibility, Trust & compliance → PUBLISH / REVISE / HOLD with the top fixes first. The page ships only on PUBLISH.
8. Deploy — Cloudflare Pages, Vercel, Netlify or GitHub Pages recipes, with crawler and schema checks after

12 modes, one command

• /page brief · /page build · /page design · /page copy · /page geo · /page gate · /page audit · /page report · /page proposal · /page compare · /page fix · /page deploy

Built for people who sell pages

• /page audit scores any live URL 0–100 across six weighted dimensions (first impression, conversion architecture, copy & proof, SEO + GEO, performance & accessibility, trust) with evidence lines, a prioritised plan with effort and estimated points, and a 30/90-day projection. /page report turns it into a white-label, print-ready client report: your logo, colours and footer from one JSON file. /page proposal writes the three-tier offer from the audit. /page compare re-audits a month later and shows the client exactly how many points they gained. /page fix applies the plan to the codebase, one commit per finding.

Who it is for

• Freelancers and agencies who build or sell landing pages and want a repeatable, defensible process with client deliverables
• Founders and marketers who need a page that converts and shows up in AI answers, not a template
• Developers who let coding agents write their front end and are tired of the AI look

What's inside

A root SKILL.md router that loads only the playbook each mode needs (small context, 12 playbooks). Five subagent definitions. Five Python scripts, standard library only, that never install packages, never read environment variables and only talk to the URL you give: an evidence collector that tests AI-crawler access, a tell checker with 62 design and copy tells (English and Italian clichés included), a JSON-LD generator, an llms.txt generator and a white-label report builder. References: the tells catalogue with fixes, a copy rulebook with a claims policy and an honest-substitute ladder for when proof is thin, a section library with eight page sequences, a CRO checklist, a GEO citability reference with the crawler allow-list, performance and accessibility budgets, a typography and colour method with pairing starters by category, and the scoring model. Templates for the brief, llms.txt, the proposal and six JSON-LD types. Worked examples.

Works with zero installs: the core runs on your agent's fetch and your editor. Python is optional and adds the scripts; Chrome adds Lighthouse and the PDF export. Works with Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot, Windsurf, OpenClaw, Cline, Zed and any other agent that reads SKILL.md.

$25, one-time, updates included. Installs in thirty seconds. Every claim in this listing describes what the package does; scores are evidence-based estimates, not guarantees of traffic or revenue.