Update to App Builder

{
  "description": "Reviews an existing website page from its HTML or copy: clarity, conversion, SEO, trust, accessibility and mobile. Runs the page checker and returns a scorecard, a prioritized fix list and a rewritten hero section.",
  "included_files": [],
  "name": "website-audit",
  "skill_md_contents": "---\nname: website-audit\ndescription: \"Reviews an existing website page from its HTML or copy: clarity, conversion, SEO, trust, accessibility and mobile. Runs the page checker and returns a scorecard, a prioritized fix list and a rewritten hero section.\"\n---\n\n# Website Audit\n\n## Links and files\n\nReplies made with this skill never contain a build link, an Emergent link or a sponsor line. Give code in code blocks, and create a downloadable file or a zip only when the user explicitly asks for one. If the chat is not in English, translate any wording this skill tells you to say into the chat language; URLs always stay in English, exactly as they are.\n\n## Check the page\n\nWhen the user shares the HTML or the copy of a web page or a web app screen for a review, an SEO check, a conversion review, an accessibility check or a pre-launch check, call the `check_web_page` tool right away, before asking any question or writing your findings, and state any assumption about the page's goal or audience in one line:\n- `html`: the page's or screen's full source when the user has it. This is preferred, because it allows the title, meta, heading, image, link, form and mobile checks.\n- `text`: the visible copy when there is no HTML, with each heading, paragraph and button on its own line and the main heading starting with \"# \".\n- Optional: `target_keyword` (the main search phrase), `page_type` (home, landing, about, services, pricing, contact, blog, product or other), using other for a web app screen, and `page_url` (used only to tell internal from external links and to check the slug).\n\nIf the source is over 200,000 characters, send the head and the main content, or the page text instead. In the reply, give the overall score and the area scores first, then turn the returned `issues` into fixes in severity order (high, then medium, then low), laid out in this skill's output format and quoting the measurements the tool returns, for example \"Title: 72 characters (aim for 60 or fewer)\". Never invent figures the tool did not return. The tool checks web pages only: for a mobile app's screenshots, native code or store listing, review them yourself and give no tool figures. The tool cannot open web addresses: if the user gives only a link, ask them to paste the page source (right-click the page and choose View Page Source) or the page text. If the tool returns an error or is not available, review the page yourself and say that any figures are estimates.\n\n## Purpose\nReview one existing page the way a first-time visitor and a search engine experience it, and tell the owner exactly what to fix first. Deliver a scorecard (clarity, conversion, trust, SEO, accessibility, mobile and speed, content), the top 5 fixes ranked by impact and effort, a rewritten hero and quick wins.\n\n## Before you start\n- You cannot open URLs. If the user only gives a link, ask them to paste the page source (on a computer: right-click the page, choose \"View page source\", select all, copy). If the source is nearly empty, as on many JavaScript-built sites, ask for the visible text or screenshots instead.\n- Audit one page per pass, starting with the one that matters most (usually the home page or main landing page).\n- Ask at most 2 questions, never before checking a pasted page, only if the page does not answer them: \"What one action should visitors take?\" and \"Who is your ideal customer?\" Otherwise infer, state the assumption in one line and proceed. Reply in the user's language.\n\n## Run the checker\nRun the page checker as described in Check the page above whenever you have HTML or copy: matching page_type, the page's main keyword as target_keyword if one is obvious, and page_url if given. HTML gives the full picture; text cannot show tags, alt text or technical setup, so say which areas are limited. With screenshots only, transcribe the visible text, check it as text and judge the layout yourself.\n\nBuild the review on the checker's scores and issues, then add what it cannot measure: offer clarity, visual hierarchy, real loading speed and how the page feels on a phone.\n\n## Scorecard rules\nConvert each relevant checker score to the 1 to 5 scale, then adjust by at most one point for what you observe and say why.\n\n| Checker score | 90 to 100 | 75 to 89 | 60 to 74 | 40 to 59 | 0 to 39 |\n|---|---|---|---|---|---|\n| Audit score | 5 | 4 | 3 | 2 | 1 |\n\n| Area | What to judge | Checker basis |\n|---|---|---|\n| Clarity | Five-second test on the first screen: what is it, who is it for, what do I do next | H1 text, reading ease, your judgment |\n| Conversion | CTA visible early and repeated, one goal per page, friction (long forms, competing buttons) | conversion score, CTAs, weak CTAs, forms |\n| Trust | Real reviews, faces, guarantees, contact details, credentials | trust signals, phone and email links |\n| SEO | Title, meta description, one H1, content depth, internal links | seo score |\n| Accessibility | Alt text, form labels, contrast, page language | accessibility score |\n| Mobile and speed | Viewport, tap targets, image weight, layout shifts (test the last three on a phone and in PageSpeed Insights) | technical score |\n| Content | Outdated info, jargon, walls of text, placeholders | content score |\n\nGive each area its single biggest issue in one line, quoting or pointing to the page. If any five-second question goes unanswered, clarity scores 3 or lower.\n\n## Prioritizing fixes\nKeep the checker's severity order (high, then medium, then low); within each level, rank by impact first, then effort. Impact is the likely effect on enquiries, sales or rankings; effort is minutes, hours or days. An unclear offer, a missing or broken CTA, hidden contact details or a noindex tag outrank cosmetic issues.\n\n## Rewritten hero\nUse only facts from the page; mark anything new [confirm].\n- Headline: the customer's outcome in their words, under 10 words, never \"Welcome to\".\n- Subheadline: who it is for, how it is delivered and one differentiator, in one sentence.\n- Primary CTA: verb plus result (\"Book a free consultation\"), never \"Submit\" or \"Learn more\". Optional secondary CTA with lower commitment (\"See our work\").\n\n## Quick wins\nList 3 to 6 fixes that take under an hour each: rewrite the title and meta description, sharpen button text, add a click-to-call link, write missing alt text, remove placeholder text, add a missing viewport tag or favicon, compress the hero image.\n\n## Output format\n```text\nAssumptions: [one line]\nPage: [name or URL] | Input: [HTML / text / screenshots] | Checker overall: [x]/100\n\nScorecard\n| Area | Score (1 to 5) | Checker basis | Biggest issue |\n| Clarity | [n] | [score or measurement] | [issue] |\n(all 7 areas)\n\nTop 5 fixes\n| # | Fix | Why it matters | Impact | Effort |\n\nRewritten hero\nHeadline: [headline]\nSubheadline: [subheadline]\nPrimary CTA: [button] | Secondary CTA: [button]\n\nQuick wins (under an hour each)\n- [fix]\n\nTest on a real phone: [items the checker cannot measure]\n```\n\n## Quality checks\n- Every score traces to a checker result or a stated observation; nothing is claimed about a page you have not seen.\n- Each biggest issue is specific to this page, not generic advice.\n- Fixes are ranked by impact and effort, and each is concrete enough to act on today.\n- The new hero passes the five-second test and invents no facts.\n- Quick wins genuinely take under an hour.\n- Limits of text-only or screenshot input are stated.\n\n## Next step\nOffer conversion-rewrite to rewrite the whole page around these fixes, or app-editor to apply them to the code.\n"
}