Visual Review logo

Visual Review

Community
dsifry
visual-review

Take screenshots of web pages and UI using Playwright for visual review and iteration

Overview

Publisherdsifry
Repositorymetaswarm
Skill namevisual-review
Stars
421
Forks
54
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by dsifry on GitHub. Read the source before you install it.

Installation

Install the Visual Review AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/dsifry/metaswarm/tree/main/skills/visual-review
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/dsifry/metaswarm.git /tmp/metaswarm
mkdir -p .claude/skills
cp -r /tmp/metaswarm/skills/visual-review .claude/skills/visual-review
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Visual Review in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Visual Review on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Visual Review is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Visual Review

Purpose

Agents cannot see rendered web pages, presentations, or UI. This skill bridges that gap by using Playwright to capture screenshots of web pages (local files, localhost servers, or deployed URLs) so agents can visually inspect their work, identify layout issues, and iterate.

This is especially useful for:

  • Reveal.js presentations — capture each slide individually
  • Web UIs — verify layout, styling, and responsiveness
  • Landing pages — check visual design matches intent
  • Email templates — verify rendering

Prerequisites

This skill requires Playwright to be available. Before first use, check and install if needed:

bash
# Check if Playwright is available
npx playwright --version 2>/dev/null

# If not available, install it (one-time setup)
npx playwright install chromium

Important: Only the chromium browser is needed. Do not install all browsers — it wastes disk space and time.

If npx playwright fails entirely, the user may need to install it:

bash
npm install -g playwright
npx playwright install chromium

Workflow

Phase 1: Setup

  1. Verify Playwright is available — run npx playwright --version
  2. If not installed — run npx playwright install chromium and inform the user
  3. Create output directory — mkdir -p /tmp/visual-review
  4. Determine the target — local file path, localhost URL, or deployed URL

Phase 2: Capture Screenshots

For a single page:
bash
npx playwright screenshot \
  --browser chromium \
  --viewport-size "1456,816" \
  --wait-for-timeout 3000 \
  "<URL_OR_FILE_PATH>" \
  /tmp/visual-review/page.png

Parameters:

  • --viewport-size "1456,816" — 16:9 aspect ratio, good for presentations
  • --wait-for-timeout 3000 — wait 3 seconds for fonts, images, and animations to settle
  • Use file:///absolute/path/to/file.html for local files
  • Use full URLs for deployed sites
For Reveal.js presentations (multiple slides):

Capture each slide by appending the slide hash to the URL:

bash
# Capture all slides (adjust count as needed)
for i in $(seq 0 15); do
  npx playwright screenshot \
    --browser chromium \
    --viewport-size "1456,816" \
    --wait-for-timeout 2000 \
    "<BASE_URL>#/$i" \
    "/tmp/visual-review/slide-$i.png"
done

Tip: To capture slides with all fragments visible, add ?fragments=true or use JavaScript injection via a Playwright script. The simpler approach is to accept that fragment-heavy slides will appear in their initial (pre-click) state.

For responsive testing:
bash
# Mobile
npx playwright screenshot --viewport-size "390,844" "<URL>" /tmp/visual-review/mobile.png

# Tablet
npx playwright screenshot --viewport-size "768,1024" "<URL>" /tmp/visual-review/tablet.png

# Desktop wide
npx playwright screenshot --viewport-size "1920,1080" "<URL>" /tmp/visual-review/desktop.png

Phase 3: Show and Review Screenshots

Use the Read tool to analyze each screenshot:

Read /tmp/visual-review/slide-0.png
Read /tmp/visual-review/slide-1.png
...

Showing screenshots to the user: The agent can see screenshots via Read, but the user cannot. When the user wants to see what you see — or when you want to discuss a visual issue together — there are two approaches depending on the environment:

Local machine (has a display):

bash
# Open a specific screenshot
open /tmp/visual-review/slide-2.png

# Open all screenshots at once
open /tmp/visual-review/slide-*.png

Remote/headless machine (SSH, tmux, etc.):

Serve the screenshots directory over HTTP so the user can view them in their local browser:

bash
# Start a simple file server (runs in background)
# Use a high port (>10000) to avoid conflicts with dev tools
python3 -m http.server 18080 --directory /tmp/visual-review &

# User can now browse to:
#   http://<hostname>:18080/
#   http://<hostname>:18080/slide-2.png
#   etc.

Pick any open high-numbered port (18080, 19090, etc.) — low ports like 8080 are often taken by dev tools. This serves the entire /tmp/visual-review/ directory with an auto-generated file listing. The user can click through all screenshots in their browser. Stop the server when done:

bash
# Stop the background server
kill %1
# Or find and kill by port
lsof -ti:18080 | xargs kill

Use whichever approach fits the environment. This is useful when collaborating on visual issues, when the user asks what something looks like, or when you want confirmation on a subjective design choice.

For each screenshot, evaluate:

  • Layout — Is content centered/aligned as intended?
  • Typography — Are headings, body text, and code readable?
  • Colors — Do accent colors, backgrounds, and contrasts look right?
  • Spacing — Is there enough whitespace? Any cramped areas?
  • Content — Is all expected content visible? (Note: fragment content may be hidden)
  • Responsiveness — Does it work at the target viewport size?

Phase 4: Document Findings

Create a findings summary:

markdown
## Visual Review — [Project Name]

### Reviewed: [date]

| Slide/Page | Status | Notes |
|-----------|--------|-------|
| slide-0   | Good   | Title centered, accent color correct |
| slide-1   | Issue  | Text overflows right edge on mobile |
| ...       | ...    | ... |

### Issues to Fix
1. [Description of issue + which slide/page]
2. [...]

### Looks Good
- [Things that are working well]

Phase 5: Fix and Re-capture

After making fixes:

  1. Re-capture only the affected slides/pages
  2. Review the new screenshots
  3. Repeat until satisfied

Reveal.js-Specific Tips

  • Slide numbering: Reveal.js uses #/0 for the first slide, #/1 for the second, etc.
  • Vertical slides: Use #/horizontal/vertical format (e.g., #/2/1)
  • Fragments: Content with class="fragment" is hidden until clicked. Screenshots show the initial state. This is expected behavior, not a bug.
  • Speaker notes: Not visible in screenshots (they appear in a separate window with S key)
  • Theme loading: Use --wait-for-timeout 3000 to ensure CDN-loaded themes and fonts are ready

Common Mistakes

MistakeWhy It's WrongDo This Instead
Not waiting for page loadFonts/images may not renderUse --wait-for-timeout 3000
Using http:// for local filesFile URLs need file:/// protocolUse file:///absolute/path
Installing all Playwright browsersWastes time and diskOnly install chromium
Treating fragment-hidden content as a bugFragments are designed to reveal on clickNote it as expected
Forgetting to create output directoryScreenshots fail silentlyAlways mkdir -p /tmp/visual-review first
Taking too many screenshots at onceFloods context windowReview in batches of 4-6
Not offering to open when discussing visualsUser can't see what you're describingUse open when collaborating on visual issues

Integration Points

Upstream: Any skill that produces visual output (presentations, web UIs, email templates)

Downstream: Fixes feed back into the source files, then re-capture to verify

Verification Checklist

  • Playwright is installed and working
  • Screenshots captured at correct viewport size
  • All pages/slides reviewed
  • Issues documented with specific descriptions
  • Fixes verified with re-captured screenshots

Frequently asked questions

What does the Visual Review AI skill do?

Take screenshots of web pages and UI using Playwright for visual review and iteration

Why use Visual Review on TypingMind?

Because you install it once and use it with any model. Visual Review is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Visual Review in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/dsifry/metaswarm/tree/main/skills/visual-review. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Visual Review?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Visual Review?

As many as you like. As long as a model supports skills, you can use Visual Review with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Visual Review AI skill free?

Yes. It is published on GitHub by dsifry under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇