pitch-deck
Make an investor pitch deck (startup or VC fund) as one HTML file from a one-line prompt, at top-seed-deck quality. Covers finding the company's design system, headline rules, the slide grid, spacing, logos and visual clarity. Use for pitch decks, seed decks, fund decks, investor slides.
- License 许可
- Apache-2.0
- 所属仓库
- getedgehq/skills
- 最近更新
- 2026-11-12
- Stars
- ★ 5
- Forks
- 0
01 前言 Front Matter
- name
- pitch-deck
- description
- Make an investor pitch deck (startup or VC fund) as one HTML file from a one-line prompt, at top-seed-deck quality. Covers finding the company's design system, headline rules, the slide grid, spacing, logos and visual clarity. Use for pitch decks, seed decks, fund decks, investor slides.
02 资源文件 2
- 参考文档DERIVATION.json1.5 KB
- 许可证LICENSE10.5 KB
03 SKILL.md 正文
pitch-deck
The bar: a partner flips through it in 60 seconds and gets the whole argument from the headlines and the
pictures alone. Work in this order, in one pass: gather → facts → headlines → design system → one
picture per slide → build → check. One Write of deck.html, Read it back once, fix with Edit.
0. Gather real material first (half of a great deck is real material, not design)
Before writing anything, collect what is real. Generic boxes are what you draw when you know nothing.
- If the prompt names a site, URL, repo or file, read it (WebFetch, Read). If it names a real company you
can look up, fetch its homepage. Take from it: the exact product name and one-liner, what the product
literally does on screen (the command, the screen, the output), the integrations or customers it names,
its real colours, fonts and logo shape (look at the CSS:
--variables,font-family, theme colours). - The brand lives in the stylesheet, not the page text (WebFetch drops
<link>tags, so never guess a CSS path). With a shell, run exactly:curl -sL <site> | grep -oiE '<link[^>]+stylesheet[^>]*>|<link[^>]+icon[^>]*>|fonts.googleapis[^"]*'then for each stylesheetcurl -sL <css-url> | grep -oE -- '--[a-zA-Z0-9-]+:[^;]{2,60}|font-family:[^;]{2,80}' | head -80. Without a shell, WebFetch the CSS URLs found in the page source. Map the variables onto §3 tokens (paper/background, surface, ink/text, muted, brand/accent colour, accent-soft) and write the mapping with its source URL in the brand comment. Never guess a palette for a company whose site you could read. - Brand = what the homepage renders, not every variable in the CSS. Stylesheets carry dozens of
colours (dark mode, unused themes, third-party widgets); the brand is what a visitor sees. Decide it from:
the page background (
<body>/<html>), the colour of the main call-to-action button in the hero ("Try", "Get started", "Sign up", "Start"), the hero headline's colour and font, and<meta name=theme-color>. If a headless browser is available (python3 -c 'import playwright'succeeds), read these from the render:python3 - <<'PY' from playwright.sync_api import sync_playwright J="()=>{const c=e=>getComputedStyle(e);return {bg:c(document.body).backgroundColor,theme:(document.querySelector('meta[name=theme-color]')||{}).content,cta:[...document.querySelectorAll('a,button')].filter(e=>{const b=e.getBoundingClientRect();return b.top<900&&b.width>60&&b.height>28&&c(e).backgroundColor!='rgba(0, 0, 0, 0)'}).slice(0,5).map(e=>[e.innerText.trim().slice(0,20),c(e).backgroundColor,c(e).color]),h1:[...document.querySelectorAll('h1,h2')].slice(0,2).map(e=>[c(e).color,c(e).fontFamily.slice(0,40)])}}" with sync_playwright() as p: b=p.chromium.launch();g=b.new_page(viewport={'width':1440,'height':900});g.goto('<site>');g.wait_for_timeout(2500);print(g.evaluate(J));b.close() PYWithout one, find the hero CTA in the HTML (curl -sL <site> | grep -oiE '<(a|button)[^>]*>[^<]{2,30}' | head -20) and resolve its class or inline colour in the CSS. paper = page background; accent = the CTA colour when it is saturated, else the most saturated colour used on the hero (headline highlight, logo); display font = the hero headline's font. Write each token with where it came from (accent #2f6bff = "Get started" button). - Real brand beats invented brand: when the company has a site, its deck uses the site's own palette, fonts and tone. §3 is only for companies with no site.
- Write the material as an HTML comment
<!-- material: ... -->at the top of deck.html: product verbs, named tools/partners, the on-screen moment to draw, brand tokens with their source. - The product slide draws THAT on-screen moment as a faithful HTML mock (the real command, the real output words), not a generic dashboard.
- Logos are the third brand win (with fonts and colours): real marks, never plain names. Harvest them
from the site itself first, because the site already ships the exact files:
curl -sL <site> | grep -oiE '<img[^>]+>' | grep -viE 'avatar|photo|\.jpe?g' | head -60For each logo note its file URL (make it absolute:<site origin>/<src>), its alt text, and the label the site puts next to it ("Works with", "Used at", "Backed by", "Customers"). Copy that label exactly; never upgrade it (a "used at" row is not "backed by"). Also take the company's own mark: the header logo or the favicon SVG from the<link rel=icon>you already found, and use that file as<img>in the footer and cover instead of redrawing it. - Named tools the site has no file for:
https://cdn.jsdelivr.net/npm/simple-icons@13.21.0/icons/<slug>.svg(lowercase slug, e.g. claude, anthropic, openai, github, slack, notion, stripe). - Verify every logo URL before using it:
curl -s -A Mozilla/5.0 -o /dev/null -w '%{http_code} ' <url>; only 200s go in. A name with no verified file is written as a wordmark in the ink colour. Write the verified list in the material comment (logos: Name=url (label)). - Show the logos where they argue: the cover row (P1), and on the product slide wherever the product meets those tools (the session window's app icon, the "runs in" line). 3-6 logos, 2.2-3cqw tall, one row, never recoloured into the accent.
- Real imagery beats drawn imagery. From the same
<img>list, also keep the site's large product screenshots, hero art and illustrations (png/webp/jpg/svg, not avatars or tiny icons; check withcurl -sIthat each is 200 and over ~20 KB). Use one where it is the truth of a slide: the product slide shows the real screen as a framed picture, the cover or close can carry the hero art. Never more than one photo-like image per slide. Skip images that only read on the website: faint, animated or mostly-white SVG backgrounds vanish at slide size; then draw the picture instead. Write them in the material comment (images: url (what it shows)). - Facts still come only from the prompt and the fetched pages; never fill gaps from memory.
1. Facts
- Use only the numbers, names and claims in the prompt. Never invent a raise, use of funds, runway, customers, logos, quotes, team, dates, competitors, market size, growth history or time savings.
- Derived arithmetic is allowed if the inputs are shown (
$120k ARR = $10k MRR × 12). - No ask given → the last slide says "Raising our seed round." (fund: no ask line at all).
- No universal words ("every", "all", "only") beyond what the facts cover. A missing number beats a weak one.
- Never print an email, a person's name or anything from the environment unless it is in the prompt.
- Claims vs. product mocks. Headlines, ledes, captions, diagram labels and bottom lines use only the
prompt's numbers: never "36 months average", "6-month rollout", "50-person team", "14.2% loss".
A product mock (a drawn screen of the product: a table, a feed, an inbox) may show realistic sample
rows ("ACH 10/24 — $4,180.00", "Ethiopia Yirgacheffe · 220 kg") because that is what the product
displays. Every mock is written exactly as
<div class="mock" data-mock> …rows… <div class="cap" style="font-size:1.2cqw">example data</div></div>(thedata-mockattribute is mandatory; a checker reads it). Nothing outside adata-mockelement may contain an unsourced number, and that includes IDs and codes ("SKU #4471", "ETH-0922", "row 14", "10/24"): outside a mock, name things in words.
2. Headlines (the deck IS the headlines)
Before any HTML, write the headlines as a list and fix them:
- Each is a full-sentence claim with a contrast or a number, disputable, max ~60 characters per line, max 2 lines. Patterns: old way. new way. / X does A. Y does B. / number + what it means. Pattern examples (fictional, shape only, never reuse the words): "It's not a model problem. It's an ingestion problem." · "Clinics book by phone. Ours books itself between visits." · "With Tally, a team closes the month in 2 days. Without it, 11."
- Never a topic label ("Problem", "Our solution", "Traction", "Team").
- Headline-only test: read them in a row as one paragraph. It must argue the whole raise.
- Name the one belief an investor holds that kills this deal ("roaster software is a tiny niche") and make sure one slide breaks it.
- After the product, every slide argues why this gets bigger (a loop, a network, a market), never more features.
- Self-score each headline 0-10 on: big, contrarian, first-principles, simple. Rewrite anything under 8.
- Open with the shift the company rides, never by insulting the reader's current tools.
Default 5-slide arc: shift/claim → gap (the structural reason it is unsolved) → mechanism (how this company closes it, next to the old way) → proof (the given numbers) → close. The close is one sentence on why it compounds plus the ask, in the §7 close layout, with no diagram. A loop diagram gets its own slide only in decks of 6+ slides; never put a diagram and the ask on one slide. Fund: thesis → what founders lack → how the fund works → portfolio facts → close (what a founder gets). More slides: add loop, product (real UI mock), traction, model, team, between proof and close.
3. Design system: find it in the company's world
The brand decides whether the deck looks like a real company in 5 seconds. Decide it before any slide. If the prompt gives a URL or a real company you know, use their real fonts, colours and logo shape. Otherwise invent the brand this company would have, from ITS world, not from a deck template.
First write an HTML comment <!-- brand: ... --> answering in one line each: what does this company's
truck / packaging / invoice / office wall look like? Which 3 materials or objects belong to its trade?
Then derive these and nothing else. A company with a real site keeps its real fonts and colours exactly
(even Inter, even a plain palette): the lists and bans below apply only when you invent a brand.
- Display face with character (headlines, cover, KPI numbers) from Google Fonts, picked by the trade:
- industrial, hardware, climate, logistics: Archivo (wdth 112-125), Barlow Condensed, Familjen Grotesk, Chivo
- finance, accounting, legal-ish B2B: Newsreader, Libre Caslon Display, Fraunces (bookkeeping heritage), or a grotesk with a distinctive cut (Archivo Narrow, IBM Plex Sans Condensed)
- infra, security, data, dev tools: Geist, JetBrains Mono for display, Schibsted Grotesk, Space Mono A neutral grotesk on a neutral ground is a template, whatever its name: give the brand one thing a person would remember (the serif, the ledger-rule texture, the stamp, the condensed caps).
- food, craft, consumer goods, hospitality: Bricolage Grotesque, Fraunces, Young Serif, Roboto Slab
- bio, health, science: Newsreader, Source Serif 4, Literata
- VC fund, editorial, legal: Instrument Serif, Gloock, Libre Caslon Display
- marketplaces, social, playful consumer: Unbounded, Rubik, Darker Grotesque
Never use Inter, Manrope, IBM Plex Sans, Instrument Sans, Figtree, Poppins or Space Grotesk as the
display face. Text face (ledes, labels, mocks) may be a plain sans that pairs with it. Set
--display,--sans, and--hw(headline weight that suits the face, e.g. 700 for condensed, 400 for serifs).
- Ground from the world: light, dark or saturated. Kraft brown for a roaster, bank navy for finance,
forest or night for climate, lab white for bio. Cream/off-white only if the trade really lives on paper.
Define
--paper(ground),--surface(panels, one step lighter/darker),--fold(muted panels),--ink,--muted(both with the ground's hue, contrast ≥ 4.5:1 on paper). - One accent (
--accent,--accent-soft) from the trade's materials. It marks only the new/winning element and the key phrase. The old way is always ink or muted. - One shape language, used on every slide: corners (0 for industrial/finance, 4px for tools, 999px for consumer), one line weight, one CSS-only texture or marker (ledger rules via repeating-linear-gradient, a stamp circle, contour lines, a grid dot). Cover object, units and panels all follow it.
- No borders anywhere. Depth = fill steps; on light grounds add one soft shadow
0 1px 2px rgba(0,0,0,.06), 0 14px 34px -22px rgba(0,0,0,.28); on dark grounds no shadow. - No gradients on type, no glow, no emoji, no icon fonts, no stock-icon cards.
- Swap test: if this deck's fonts and colours could belong to any other startup, choose again.
4. The slide system (copy this CSS verbatim, then add the parts you use)
Everything is in cqw (1cqw = 19.2px at 1920 wide). This is the grammar of a deck that raised a top
seed round; only the tokens from §3 change per company.
*{box-sizing:border-box}
body{margin:0;background:var(--paper);color:var(--ink);font-family:var(--sans);-webkit-font-smoothing:antialiased}
.slide{position:relative;width:100%;aspect-ratio:16/9;container-type:inline-size;overflow:hidden;background:var(--paper)}
.in{position:absolute;inset:0;padding:5.4cqw 5.4cqw 7cqw;display:flex;flex-direction:column}
.foot{position:absolute;left:5.4cqw;right:5.4cqw;bottom:2.6cqw;display:flex;justify-content:space-between;align-items:center;font-size:1.2cqw;color:var(--muted)}
.foot .mk{display:flex;align-items:center;gap:.6cqw;color:var(--ink);font-family:var(--display);font-size:1.45cqw;letter-spacing:-.03em}
.label{font-size:1.2cqw;color:var(--muted);margin:0 0 1.4cqw;letter-spacing:.01em}
h1,h2,h3{font-family:var(--display);font-weight:var(--hw);margin:0;text-wrap:balance;color:var(--ink)}
h1{font-size:6.4cqw;line-height:1.03;letter-spacing:-.045em;max-width:12.5ch}
h2{font-size:4.1cqw;line-height:1.07;letter-spacing:-.04em;max-width:26ch}
p{margin:0}
.lede{font-size:1.75cqw;line-height:1.4;color:var(--muted);max-width:42ch;margin-top:1.8cqw;text-wrap:pretty}
.a{color:var(--accent)}
.stage{flex:1;min-height:0;display:flex;align-items:center;margin-top:2.4cqw}
.stage>svg{display:block;width:100%;height:auto;max-height:100%}
.card{background:var(--surface);border-radius:1.2cqw;box-shadow:var(--shadow)}
.num{font-family:var(--display);font-variant-numeric:tabular-nums;letter-spacing:-.055em;line-height:1}
--shadow: 0 1px 2px rgba(0,0,0,.06),0 14px 34px -22px rgba(0,0,0,.28) on light grounds, none on dark.
Type scale, never smaller: label/foot/captions 1.2cqw (23px), body 1.45cqw, sub 1.75cqw, h2 4.1cqw,
cover h1 6.4cqw, big numbers 5cqw. Every content slide: label + h2 (+ optional one-line lede) top-left,
ONE part in .stage, footer = wordmark left, page number right. No borders anywhere.
5. Parts (pick one per slide; never the same part twice; fill with this company's real words)
P1 · cover
<section class="slide s-cover"><div class="in" style="justify-content:center;padding-bottom:5.4cqw">
<h1>{the shift as a claim, ≤ 8 words}</h1>
<p class="lede" style="max-width:30ch">{what the company is, one sentence}</p>
<div style="display:flex;align-items:center;gap:1.6cqw;margin-top:4cqw;flex-wrap:wrap">
<span style="font-size:1.2cqw;color:var(--muted)">{Works with | Built for | Backed by}</span>
{3-6 × <span style="display:inline-flex;align-items:center;gap:.6cqw;font-size:1.3cqw"><img src="{verified logo url}" alt="{Name}" style="height:2.2cqw;width:auto">{Name}</span>}
</div>
<svg style="position:absolute;right:5.4cqw;top:50%;transform:translateY(-50%);width:36cqw;height:36cqw" viewBox="0 0 400 400">{the cover object: the claim in one drawing, ≥ 8 shapes, one accent detail}</svg>
</div><div class="foot"><span class="mk">{mark svg}{Name}</span><span>{domain or nothing}</span></div></section>
The row only lists names the prompt or the fetched site mention (integrations, customers, partners), with the site's own label and verified logo files. No names → drop the row. Keep the h1 column clear of the object (h1 max 12.5ch). The logo row is ONE line: at most 5 logos, all inside the text column, never wrapping, never under the picture or into the footer. More customers → the proof slide gets the logo wall.
P2 · fan-in: many → few (the shift, the gap, a market converging)
Left: 4-6 pills (who/what), right: 2-3 cards (where it all goes); the winning card in accent.
<div class="stage"><svg viewBox="0 0 1000 440">
<!-- pill i at y = 40 + i*72 (i = 0..5); card j at y = 110, 220, 330 -->
<!-- one curve per link: --> <path d="M290 {yi} C 500 {yi}, 520 {yj}, 700 {yj}" fill="none" stroke="var(--fold)" stroke-width="1.7"/>
<!-- pill: --> <rect x="20" y="{yi-28}" width="270" height="56" rx="28" fill="var(--surface)"/><text x="48" y="{yi+9}" font-size="25" fill="var(--ink)">{name}</text>
<!-- card: --> <rect x="700" y="{yj-38}" width="270" height="76" rx="16" fill="var(--surface)"/><text x="732" y="{yj+11}" font-size="31" fill="var(--ink)">{name}</text>
<!-- winning card: fill var(--accent), text #fff; its curves stroke var(--accent) width 2.4 -->
</svg></div>
Names ≤ 16 characters (pill text is 25 units). Use font-family = var(--sans) on the <svg>.
P3 · three columns, the last one breaks (everyone does X, here it fails)
<div class="stage"><div style="display:grid;grid-template-columns:repeat(3,1fr);gap:2.4cqw;width:100%">
<!-- column ×3; the third gets background:var(--accent-soft) and accent text -->
<div style="display:flex;flex-direction:column;gap:1.2cqw">
<p style="font-size:1.45cqw">{column head}</p>
<div style="background:var(--fold);border-radius:1.2cqw;padding:1.6cqw;display:flex;flex-direction:column;gap:1cqw;align-items:center">
<div style="display:flex;gap:.8cqw;flex-wrap:wrap;justify-content:center">{2-4 × <span style="background:var(--surface);border-radius:999px;padding:.6cqw 1.2cqw;font-size:1.25cqw">{chip}</span>}</div>
<svg width="2cqw" height="4cqw" viewBox="0 0 20 40"><path d="M10 2V36M3 29l7 8 7-8" fill="none" stroke="var(--ink)" stroke-width="2.4" stroke-linecap="round"/></svg>
<div style="background:var(--ink);color:var(--paper);border-radius:1cqw;padding:1.2cqw 2cqw;font-size:1.45cqw;width:100%;text-align:center">{where it ends}</div>
</div>
</div>
</div></div>
<p style="font-size:1.2cqw;color:var(--muted);text-align:right;margin-top:1cqw">{one lowercase caption under the breaking column}</p>
All three columns share the same rows, so the eye compares them line by line.
P4 · the product moment (what the product literally does, on screen)
<div class="stage" style="display:grid;grid-template-columns:7fr 5fr;gap:4.4cqw;align-items:center">
<div class="card" style="overflow:hidden" data-mock>
<div style="display:flex;align-items:center;gap:.5cqw;padding:1.1cqw 1.4cqw;background:var(--fold)"><i style="width:.8cqw;height:.8cqw;border-radius:50%;background:var(--paper)"></i><i style="width:.8cqw;height:.8cqw;border-radius:50%;background:var(--paper)"></i><i style="width:.8cqw;height:.8cqw;border-radius:50%;background:var(--paper)"></i><span style="margin-left:1cqw;font-size:1.2cqw;color:var(--muted)">{the real app or screen name}</span></div>
<div style="padding:1.8cqw 2cqw 2.2cqw;display:flex;flex-direction:column;gap:1.25cqw;font-size:1.25cqw;line-height:1.5">
<p style="background:var(--paper);border-radius:.8cqw;padding:.9cqw 1.2cqw;font-size:1.4cqw;align-self:flex-start">{what the user types or does}</p>
{2-4 rows: <div style="display:grid;grid-template-columns:1.4cqw 1fr;gap:.6cqw;color:var(--muted)"><span style="width:.8cqw;height:.8cqw;border-radius:50%;background:{var(--accent) for the product's own step, var(--muted) otherwise};margin-top:.6cqw"></span><div><b style="font-weight:500;color:{var(--accent)|var(--ink)}">{action}</b> {detail}</div></div>}
{the result, once: <div style="background:var(--accent-soft);border-radius:.7cqw;padding:.8cqw 1cqw;color:var(--ink)">{the output}<small style="display:block;color:var(--muted);font-size:1.2cqw">{why it is right}</small></div>}
<p style="font-size:1.2cqw;color:var(--muted)">example</p>
</div>
</div>
<div style="display:flex;flex-direction:column;gap:1.3cqw">
{3 × <p style="display:grid;grid-template-columns:2.6cqw 1fr;gap:1cqw;align-items:baseline;font-size:1.7cqw;line-height:1.35"><span style="font-size:1.2cqw;color:var(--accent)">{1|2|3}</span><span>{step in plain words}</span></p>}
</div>
</div>
The rows are the real product verbs from the gathered material (a command, a call, a click), never "Step 1: Onboarding".
P5 · before vs after in units (a result with a count: "4 of 20", "11 of 12")
<div class="stage" style="flex-direction:column;align-items:stretch;justify-content:center;gap:3.6cqw">
<!-- row ×2: first the old way (marks in var(--ink)), then the new way (marks in var(--accent)) -->
<div style="display:grid;grid-template-columns:22cqw 1fr 12cqw;gap:3cqw;align-items:center">
<p style="font-size:1.75cqw;line-height:1.3">{who}<span style="display:block;color:var(--muted);font-size:1.45cqw;margin-top:.4cqw">{condition}</span></p>
<div style="display:grid;grid-template-columns:repeat({N},1fr);gap:.7cqw">{N × <i style="aspect-ratio:1;border-radius:.5cqw;background:{lit ? colour : var(--fold)}"></i>}</div>
<p class="num" style="font-size:5cqw;text-align:right">{k}<small style="font-size:1.7cqw;color:var(--muted)">/{N}</small></p>
</div>
<!-- optional band with one more given number: -->
<div style="display:flex;align-items:baseline;gap:2cqw;background:var(--accent-soft);border-radius:1.2cqw;padding:1.6cqw 2.4cqw">
<span class="num" style="font-size:4.4cqw;color:var(--accent)">{value}</span><p style="font-size:1.75cqw">{what it means}</p></div>
</div>
N ≤ 30 per row. One row only when there is no "before" (then the lit subset is the contrast).
P6 · growth: bars + KPI column (only when the prompt gives at least a start and an end value)
<div class="stage" style="display:grid;grid-template-columns:7.6fr 4.4fr;gap:4cqw;align-items:center">
<svg viewBox="0 0 900 420"><!-- one bar per GIVEN value only; never invent the points in between.
Two values → two bars (start, now) with the multiple between them. Bar width 160, rx 12,
height ∝ value, baseline y=380; earlier bars var(--fold), the last bar var(--accent);
value above each bar at font-size 34, period label below at 24 in var(--muted). --></svg>
<div style="display:flex;flex-direction:column;gap:2.4cqw">
{2-3 × <div><div class="num" style="font-size:5.2cqw">{given or derived-with-inputs value}</div><p style="font-size:1.45cqw;color:var(--muted);margin-top:.6cqw">{what it is}</p></div>}
</div>
</div>
P7 · once vs every time (a business model, a per-use economy, one-off vs recurring)
<div class="stage" style="display:grid;grid-template-columns:1fr 1.5fr;gap:1.6cqw;align-items:stretch">
<div class="card" style="padding:2.2cqw 2.4cqw;display:flex;flex-direction:column;gap:1.4cqw">
<div class="num" style="font-size:3.4cqw">{Sells once}</div>
<div style="flex:1;display:flex;align-items:center"><div style="width:8cqw;height:8cqw;border-radius:.8cqw;background:var(--ink);color:var(--paper);display:flex;align-items:center;justify-content:center;font-size:1.2cqw">{one unit}</div></div>
<p style="font-size:1.45cqw;color:var(--muted)">{the old unit, sold once}</p></div>
<div style="background:var(--accent);color:#fff;border-radius:1.2cqw;padding:2.2cqw 2.4cqw;display:flex;flex-direction:column;gap:1.4cqw">
<div class="num" style="font-size:3.4cqw">{Sells every use}</div>
<div style="flex:1;display:grid;grid-template-columns:repeat(18,1fr);gap:.55cqw;align-content:center">{54 × <i style="aspect-ratio:1;border-radius:.35cqw;background:#fff;opacity:.9"></i>, every 3rd opacity .6}</div>
<p style="font-size:1.45cqw;opacity:.85">{the new unit, sold every time}</p></div>
</div>
P8 · close (the compounding picture, never text alone)
The close is the headline plus ONE picture of why it compounds, filling the right ~55% of the stage. Pick what fits this company: (a) a loop: 4 rounded labels on an ellipse with accent arrows between them (inline SVG, arrows in the accent, labels are this company's own nouns: users, data, supply...); (b) once vs every: a single dark tile next to an accent panel full of small tiles (one sale vs recurring use); (c) the site's real hero art. Below the headline: the ask as given, else "Raising our seed round.", then the domain.
<section class="slide"><div class="in"><div style="display:grid;grid-template-columns:42% 1fr;gap:4cqw;align-items:center;height:100%">
<div><h2 style="font-size:4.6cqw;max-width:14ch">{why it compounds, <span class="a">key phrase</span>}.</h2>
<p style="font-size:1.75cqw;color:var(--muted);margin-top:2.4cqw">{ask or "Raising our seed round."}</p></div>
<svg viewBox="0 0 800 520" style="width:100%;height:auto">{the loop or once-vs-every, labels ≥ 26px in viewBox units}</svg>
</div></div><div class="foot"><span class="mk">{mark}{Name}</span><span>{domain}</span></div></section>
P10 · old way vs new way (the gap as one big pair)
Two panels side by side filling the stage: left on --surface, the old way drawn with this company's real
artifacts (the stale docs, the manual steps, the tools that disagree) slightly faded or tilted; right on
--accent-soft, the product in the middle feeding or fixing the same things, accent on the product.
One caption line under each panel. Words only in captions (no invented counts or versions).
P9 · loop (a flywheel as a content slide, when the claim is how it compounds)
Same picture as P8(a) but full stage width: an ellipse of 4 labels ~3cqw tall, accent arrows, the headline above.
6. Which part where (default 5 slides)
1 cover → P1 · 5 close → P8. For slides 2-4 choose by what THIS company's claim is, not by habit: the shift or gap → P10 (old vs new pair), P2 (many sources into one) or P3 (three ways, the last breaks; only when the company itself names three alternatives) or a before/after of the real product; the product → P4 or the site's real screenshot framed large; proof → P5 (a count), P6 (growth), a quote with the customer's logo big, or a logo wall when customers are the proof; business model → P7; compounding → P9. Two companies with different stories should not get the same three middle slides. More slides: team, market, ask, each drawn, never a bullet list.
Scale like a top deck. The picture on every slide fills the stage: at least 85% of the content width, or at least 60% of the content height when it sits beside text. Inside pictures: labels and pills ≥ 1.4cqw, card titles ≥ 2.4cqw, big numbers ≥ 6cqw. Empty half-slides are a failure. Bigger must still fit: the stage starts below the headline and ends above the footer band; nothing in the picture may touch the headline, the footer or the slide edge (an SVG keeps its outermost labels inside its viewBox).
7. Check (Read the file back once, then Edit)
- Slide count is exactly what the prompt asked (default 5). The
<!-- material -->and<!-- brand -->comments exist. Every<img>src is an absolutehttps://URL (never a relative path likelogos/x.svg: deck.html is opened on its own) that you verified as 200; the site's logos appear on the cover and the product slide; the footer uses the company's real mark file when it has one. - Every number on a slide is in the prompt, on the fetched site, or derived with shown inputs;
product mocks carry
data-mock. - Headlines in a row argue the whole raise; each is a claim with a contrast; no topic labels.
- Each content slide uses a different part from §5 and the part fills the stage.
- Every slide's picture fills the stage (see §6 scale); the close slide has a picture. Nothing under 1.2cqw, nothing past the slide edge or into the footer band, nothing overlapping. Pictures too: a window, card or SVG must start below the headline's last line and end above the footer; when it does not fit, shrink the picture, never let it slide under the headline or footer.
- One accent, used only for the new/winning element and the key phrase. No borders.
- Same label/headline position on every content slide; footer on every slide.
04 同仓库更多 Skills
zero-downtime-cloud-move
Runbook for moving a live app from one cloud to another (Google Cloud to AWS, AWS to Google Cloud, a PaaS to your own account, or one storage provider to another) without downtime. The agent inventories what is really running, builds the new stack next to the old one, keeps a one-line rollback for every phase, proves the new stack with real workloads before any traffic moves, cuts over with a checklist, watches a fixed window after the switch, copies data with an idempotent job whose second pass must report 0 left, and deletes the old cloud last. Use when someone asks to migrate, move or replatform a production app, switch cloud providers, move a backend, database files or object storage, plan a DNS cutover, or decommission an old cloud account.
zayats-ceo-branding
Build an evidence-based CEO positioning paper, Personal Branding Roadmap, and editorial workflow using Marina Zayats's public method. Use for executive branding, LinkedIn ghostwriting briefs, profile strategy, and recurring CEO-approved content planning.
yc-app-roast
YC App Roast: roast a YC or a16z Speedrun application. Use when a founder wants their YC or Speedrun answers scored, asks "what would a YC partner think", wants the weak answers found or rewritten, or wants their draft checked before they submit. Scores every answer 0-10 against public advice from YC and Speedrun partners, names the top issue with the exact words that cost points, and writes a rewrite from the founder's own facts. Works for YC and a16z Speedrun. Not affiliated with Y Combinator or a16z.
yc-app-review
YC App Review: get a YC or a16z Speedrun application reviewed by people you trust. Use when a founder wants friends, mentors or alumni to review their YC or Speedrun application, or asks to share their draft for feedback. Builds a shareable review page (one HTML file, application-portal look) where reviewers leave notes per question and press "Send my notes" to send them back as text, then parses the notes into proposed rewrites. Optional tracked private links per reviewer. Works for YC and a16z Speedrun. Not affiliated with Y Combinator or a16z.
x-growth-playbook
X Growth Playbook. Analyse a person's own X (Twitter) posts and tell them what to post next, with the numbers behind every claim. Compares quote posts vs originals, links vs no links, short vs long, time of day in their timezone, media vs text and the topics they name, reports medians with n, then writes a 10-step plan and the next 7 posts in their voice. Use when the user says "analyse my X posts", "what works on my Twitter", "how do I grow on X", "what should I post next on X", "audit my X account", or shares an X data archive, X API export or a list of their posts. Runs on the user's computer.
workplan
Create, update, or close work plans for multi-step tasks. Use when starting refactors, bug lists, feature work, migrations, or any task with 2+ steps. Also use after auto-compaction to re-orient. Triggers: "workplan", "work plan", "create a plan", "what's the plan", "where was I", or when Claude detects a multi-step task that needs tracking.