Skip to content
cccskills
FreePublished on GitHub

create-chatgpt-mockup

Turns a thread JSON into a single still image of a ChatGPT-style iOS screen in light mode. It supports user text and image messages, assistant markdown, citation chips, and composer details, then outputs standalone HTML and a 9:16 PNG for mockups or creative work.

OriginalShow the original description

Render pixel-accurate ChatGPT mobile (iOS) screen mockups in light mode from a thread JSON. Supports user text bubbles, user image attachments, assistant markdown prose, citation chips, the OpenAI spiral logo, the Apps-SDK GPT chip in the composer, and three header styles (model-tag, plain title, "Get Plus"). Fixed 9:16 viewport. Outputs HTML + PNG.

gooseworks-ai/goose-skills1,272 stars(GitHub 1,272 + cccskills 0)

Run score Not tested

See how to install

Good for

  • Creating a ChatGPT screen for a social post
  • Showing a sample conversation in an ad creative
  • Making a product mockup from thread JSON
  • Preparing a still chat screen for a storyboard

Files included (20)

  • SKILL.md14.9 KB
  • examples/01-sunscreen-image.json467 B
  • examples/02-sleep-apnea-long.json1.1 KB
  • examples/03-poem-iphone-mac.json1007 B
  • examples/04-empty-typing.json258 B
  • examples/05-seatfinch-gpt-chip.json596 B
  • examples/06-short-howto.json750 B
  • examples/assets/sunscreen-sky.svg3.1 KB
  • generate.js16.3 KB
  • package.json348 B
  • render.js3.0 KB
  • screenshot.js2.5 KB
  • skill.meta.json136 B
  • templates/chat.css16.6 KB
  • templates/icons.js11.5 KB
  • tests/expected-output.md2.8 KB
  • tests/human-test.md1.8 KB
  • tests/run-all.sh414 B
  • tests/sample-input.md1.8 KB
  • tests/smoke-test.md796 B

SKILL.md (original)

Review the instructions the agent will receive before you install.

Superseded: the video kit now does this with the phone-chat part, version 1.2.0, in the parts folder of this repository. Every phone-chat style is a style file that draws the same screen frame by frame from the plan's scenes, with the original sounds, in the kit's browser. This atom stays, unchanged in behaviour, for skills outside the kit until they move; its scripts still run.

create-chatgpt-mockup

Purpose

Generate single-frame ChatGPT mobile screens that read as real iOS captures — accurate header chrome, user/assistant message styling, composer pill, status bar, and the OpenAI spiral mark. Designed for use inside ad creatives, social posts, video storyboards, and product mockups.

If you need an animated, video-recorded chat reveal (bubbles popping over time, SFX, music bed, end card), use the render-chatgpt-chat skill. This atom only produces a single still frame.

Inputs

Compose a thread.json matching the schema below, then pass it to render.js.

Thread JSON schema

{
  "statusBar": {
    "time": "9:41",
    "dnd": false
  },
  "header": {
    "style": "model-tag",
    "title": "ChatGPT",
    "model": "5.1",
    "rightIcons": ["personPlus", "dottedCircle"]
  },
  "messages": [
    { "type": "user-image", "src": "assets/photo.jpg", "aspect": "square" },
    { "type": "user-text",  "text": "Make this face sunscreen better" },
    {
      "type": "assistant",
      "title": "Optional H1 with spiral logo",
      "feedback": true,
      "text": "Markdown-ish prose. **bold**, *italic*, ## H2, * bullet, 1. ordered, [[cite:Source +1]], [[icon:💡]] inline-icon."
    }
  ],
  "composer": {
    "placeholder": "Ask anything",
    "text": "Typed text to show with caret",
    "cursor": true,
    "streaming": false,
    "chip": { "name": "Seatfinch" }
  }
}

Field reference

statusBar

  • time — string shown top-left (e.g. "9:41", "08:18").
  • dnd — when true, renders a small moon glyph next to the time (matches iOS Focus-on indicator).

header.style — one of:

  • "model-tag" — ≡ ChatGPT 5.1 ▾ ●● — used when a specific model is active.
  • "title-only" — ≡ ChatGPT › ●● — the plain in-chat header.
  • "plain-title" — ≡ Get Plus ✦ ●● — for the upsell/empty-state title.

header.rightIcons — array of icon keys, rendered left-to-right on the right side. Available: "personPlus", "dottedCircle", "edit", "more". Default is ["personPlus", "dottedCircle"] for model-tag, ["edit"] for title-only.

messages[] — one of:

  • { "type": "user-image", "src": <path>, "aspect": "square"? } — relative src paths resolve against the thread.json directory.
  • { "type": "user-text", "text": <string> } — right-aligned soft-gray bubble.
  • { "type": "assistant", "text": <markdown>, "title"?: <string>, "feedback"?: <bool> } — left-aligned plain prose. If title is set, an OpenAI spiral logo + bold title is rendered above the body. feedback: false hides the thumbs-up/down chips.

Assistant text markdown — supported syntax:

  • **bold**, *italic*
  • # H1, ## H2, ### H3
  • * item / - item bullet lists
  • 1. item ordered lists
  • --- horizontal rule
  • [[cite:Some Source +1]] — pill citation chip
  • [[icon:💡]] — inline emoji prefix (kept un-escaped)
  • Single \n → soft line break (<br>); double \n\n → new paragraph

composer

  • placeholder — shown when text is empty (default "Ask anything").
  • text — typed-in text. When present, the send button activates (black bg).
  • cursor — whether to render the blinking caret after text (default true).
  • streaming — when true, replaces the send arrow with a stop square (matches "response generating" state).
  • chip — optional GPT chip in the composer. chip.name is the GPT name. Sets placeholder to "Ask ChatGPT" if you don't override it.

CLI

node render.js --thread examples/01-sunscreen-image.json
node render.js --thread my-thread.json --output ./exports --name nightly

Flags

FlagEffect
--thread <path>(required) JSON file matching the schema above
--output <dir>parent dir for the dated output folder; default ./chatgpt-mockup-exports/
--name <slug>override the output folder slug

Output

<output>/<YYYY-MM-DD>-<slug>/
  index.html       # standalone HTML (CSS inlined)
  screenshot.png   # 2250 × 4002 PNG (DPR 3, 9:16)
  thread.json      # copy of the input for reproducibility

Workflow

  1. Receive a brief or existing thread JSON.
  2. If only a brief is given, compose thread.json matching the schema above.
  3. Run node render.js --thread <path>.
  4. Open the PNG. If anything looks off (header alignment, bubble width, missing icon), edit the JSON or templates/chat.css and re-render.

Setup (one-time)

cd create-chatgpt-mockup
npm install
npx playwright install chromium

Quality checks

  • Renders in light mode only (white background, dark text). Dark-mode rendering is out of scope for v1.
  • Status bar time and right cluster (signal / wifi / battery) sit at the very top, no clipping under the iOS status-bar safe area.
  • Header chrome icons (hamburger / personPlus / dottedCircle / edit / more) all visually weigh the same as a real ChatGPT iOS screenshot at the same display size — eyeball, don't pixel-match. Sub-reading-weight icons read as broken even when their pixel dimensions "match the spec."
  • Header model tag (5.1, 5.2) sits inline with ChatGPT and its chevron — no wrapping.
  • User bubble is right-aligned, soft-gray (#F4F4F4), border-radius: 28px, max-width 78% of stage.
  • User image attachments use the same 28px corner radius and sit right-aligned above the text bubble.
  • Assistant body is plain prose (no bubble), left-aligned, full-width within the gutter.
  • Composer corner radius is 999px (full pill), 1px border #E5E5E5, send button is 44 × 44 round.
  • When composer.text is set: send button is solid black with a white up-arrow. Caret sits right after the last character of typed text.
  • When composer.streaming: true: send button shows a black square stop icon.
  • When a citation marker [[cite:Foo +1]] is in the assistant text, it renders as a pill chip inline at the end of the sentence.

Failure modes

SymptomCauseFix
Thread file not foundWrong path or running from a different cwdUse absolute paths or run from the skill directory
Image attachment shows broken-image glyphsrc is relative but the renderer couldn't resolve itrender.js resolves relative src against the thread.json's directory — make sure the asset exists at that path
Send button stays light gray after typingcomposer.text is empty string "" instead of unsetEither omit text entirely or set it to a real string
Caret renders as a stray | separated from the textCaret span is being treated as a flex childThe CSS rule for .composer .input uses inline content flow — do not switch it to display: flex
Streamed list-item words render with no spaces ("Howconversationswork").word { display: inline-block } collapses inter-span text-node whitespace inside <div> parents (works fine inside <p>)Use display: inline for .word — modern browsers honor filter: blur() + opacity transitions on inline elements too
Bullets visible while list-item text is still hidden during streamingCSS list-style markers are browser-painted; the word-stream wrapper only sees text nodesThe atom auto-converts <ul><li>X</li></ul> to <div class="md-list"><div class="md-li">• X</div></div> when stream: true — keep that pass; don't regress it. Any new markdown that emits browser-painted glyphs (::marker, quotes, counter-content) needs the same text-conversion treatment.
Markdown --- doesn't render an <hr>The --- line isn't on a line by itself, or the surrounding text didn't end its paragraphSurround --- with blank lines on both sides
Stanzas of a poem joined into single paragraphsLines separated by single \n get joined as a single paragraph but with <br> between them — this is the intended behavior. If you want stanza breaks, use \n\n between stanzas.
Header right cluster overlaps the model tag on long titlesTitle text is too wide for the centered slotShorten header.title, drop the model tag, or switch header.style to "title-only"

Files

FilePurpose
render.jsCLI entry — parses flags, generates HTML, takes screenshot, writes outputs
generate.jsThread JSON → standalone HTML page (markdown-ish parser inside)
screenshot.jsHTML → PNG via Playwright (chromium headless) at fixed 750×1334 / DPR 3
templates/chat.cssAll visual styling
templates/icons.jsInline SVG icons (hamburger, chevrons, spiral logo, mic, plus, send arrow, stop square, signal/wifi/battery, thumbs up/down)
examples/*.json6 reference threads exercising every variant
examples/assets/Placeholder image assets used by image examples
tests/run-all.shBuilders only, in the goose-skills repo (installs have no tests/): render every example into tests/output/ for visual review

Known limitations

  • Light mode only. Dark mode would require a separate theme block and is out of scope for v1.
  • No carousel / horizontal-scroll cards. The Apps-SDK ticketing-app reference shows a horizontal card carousel inside the assistant response; not modeled here.
  • No live keyboard rendering. The composer is the bottom pill only — no on-screen keyboard chrome, suggestion bar, or QWERTY grid. (If you need the keyboard, scale the screen down 50% and overlay a separate keyboard image.)
  • Markdown is intentionally minimal — no tables, no fenced code blocks with syntax highlighting, no nested lists. Add to generate.js#renderMarkdown if a brief requires more.
  • Emoji rendering depends on system fonts. Chromium on macOS ships Apple Color Emoji; on Linux ships Noto Color Emoji.

Quality Checks

Render each examples/*.json with render.js and open the PNGs (builders in the goose-skills repo can run bash tests/run-all.sh, which renders all six into tests/output/), then confirm:

  • Every example renders without error. All six examples/*.json produce an index.html + screenshot.png + thread.json triplet; render.js exits 0.
  • Dimensions are stable. Each screenshot.png is 2250 × 4872 (the .stage is 750 × 1624 and screenshot.js captures at deviceScaleFactor: 3). A wrong size means the stage box or DPR changed.
  • Light mode, white background. Background is white, text near-black. This atom does not render dark mode.
  • Status bar. Time sits top-left; the signal / wifi / battery cluster sits top-right, unclipped. When statusBar.dnd: true (see 02-sleep-apnea-long.json, 05-seatfinch-gpt-chip.json) a moon glyph renders next to the time.
  • Header style matches the JSON. model-tag shows ChatGPT 5.1 ▾ (01-sunscreen-image.json, 06-short-howto.json); title-only shows the plain title + right chevron (02, 03, 05); plain-title shows the upsell title with no chevron (04-empty-typing.json). The center cluster never wraps and the right icons never overlap it.
  • User turns. User text is a right-aligned soft-gray bubble (#F4F4F4, 28px radius). A user-image (see 01-sunscreen-image.json) renders the attachment right-aligned, same 28px corners, above the text bubble.
  • Assistant turns. Assistant prose is left-aligned plain text with no bubble. Markdown renders: bold/italic, ## headings, bullet and ordered lists (06-short-howto.json), --- horizontal rules and the [[cite:…]] pill chip (02-sleep-apnea-long.json), an [[icon:💡]] inline prefix, and title as bold heading text (03-poem-iphone-mac.json, 05-seatfinch-gpt-chip.json). feedback: false hides the thumbs chips.
  • Empty state. When messages is empty (04-empty-typing.json) the centered OpenAI spiral hero renders instead of a conversation.
  • Composer. Default is the full-pill (999px) bar with +, input, mic, and a round 44×44 send button. With composer.text set (04-empty-typing.json) the send button is solid black and a caret follows the typed text. With streaming: true (02, 03) the send button shows the black stop square. With composer.chip (05-seatfinch-gpt-chip.json) the Apps-SDK GPT chip renders inside the composer above the input row.

Failure Modes

SymptomLikely causeWhat to do
Thread file not found: … then exit 1--thread path is wrong or the cwd is not the skill dirPass an absolute path or cd into the skill directory first (render.js resolves the path against cwd)
Error: playwright not installednpm install never ran in the skill dirRun npm install, then npx playwright install chromium (see Setup)
Run hangs ~15s then page.goto … timeoutscreenshot.js is waiting on networkidle for an asset that never loads (e.g. a remote/http image src)Use a local relative src that resolves under the thread.json dir, or an inline SVG asset like examples/assets/sunscreen-sky.svg
Image attachment shows the broken-image glyphuser-image.src is relative but the file is missing on diskrender.js rewrites a relative src to file://<threadDir>/<src>; make sure the asset actually exists there
Send button stays gray after setting textcomposer.text is "" (empty string) rather than a real valueSet composer.text to a non-empty string, or omit it to keep the placeholder state
PNG comes out the wrong size.stage width/height in templates/chat.css or DEVICE_SCALE/VIEWPORT in screenshot.js was changedKeep .stage at 750 × 1624 and deviceScaleFactor at 3 so output stays 2250 × 4872
Header right icons overlap the model tag on a long titleheader.title is too wide for the centered slotShorten the title, drop the model tag, or switch header.style to "title-only"
[[cite:…]] or [[icon:…]] prints literally instead of renderingThe marker is mistyped (missing double brackets / colon)Match the exact [[cite:Source +1]] / [[icon:💡]] form parsed by formatInline in generate.js
Markdown --- shows as text, not an <hr>The --- line is not on its own line between blank linesPut a blank line above and below the ---, as in 02-sleep-apnea-long.json

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

<slug>

FreeSummary

The supplied material contains only placeholder text for the skill’s purpose, inputs, steps, and completion criteria. It does not provide enough information to describe a task or result reliably.

  • No use case can be identified from the supplied material
  • Purpose is unspecified
  • Inputs and expected results are unspecified
gooseworks-ai/goose-skills1.3KUpdated Oct 27, 2026

ad-angle-miner

FreeSummary

Research supported advertising angles using buyer context, current product facts, customer language, and observed ads. Turn the evidence into static ad, video, or copy ideas, matching video concepts to available recipes. Scores rank ideas for testing; they do not predict results.

  • Finding evidence-backed angles for a product campaign
  • Developing static ad or video concepts from customer research
  • Writing headline and primary-text options for ads
gooseworks-ai/goose-skills1.3KUpdated Oct 27, 2026

Analyze campaign data from Google, Meta, and LinkedIn to find strong and wasteful spend. The skill checks funnel performance and test results, then recommends which campaigns to pause, scale, or test and how to shift budget across channels.

  • Finding campaigns and ads that waste budget
  • Comparing performance across Google, Meta, and LinkedIn
  • Planning specific budget shifts and follow-up tests
gooseworks-ai/goose-skills1.3KUpdated Oct 27, 2026

Compare ad spend and signups with downstream outcomes to calculate the cost of each qualified lead. The skill checks tracking coverage, accounts for how long leads need to mature, and groups creatives into actionable categories with confidence notes.

  • Compare true qualified-lead costs across creatives
  • Find low-CPA ads that bring poor-quality leads
  • Check whether signups can be linked to downstream outcomes
gooseworks-ai/goose-skills1.3KUpdated Oct 27, 2026

Prepare the intake for a Meta ad campaign by reviewing available brand information first, stating reasonable assumptions, and asking only questions whose answers could affect budget, claims, or legal exposure. It records stable brand facts, a campaign brief, and what was asked versus assumed.

  • Prepare a Meta ad campaign when brand facts are missing
  • Clarify budget and timing before campaign planning
  • Record approved claims and audience restrictions
gooseworks-ai/goose-skills1.3KUpdated Oct 27, 2026

ad-policy-check

FreeSummary

Review ad copy, image text, visuals, and people against platform and brand category rules. It suggests honest fixes and needed disclosures, then flags issues requiring human or legal review, with a separate result for each ad and platform.

  • Checking ad copy and visuals before launch
  • Reviewing creator ads for required disclosures
  • Fixing a platform disapproval and recording the issue
gooseworks-ai/goose-skills1.3KUpdated Oct 27, 2026

See all skills by gooseworks-ai

Report a problem with this skill