PR Screenshot Skill

Add before/after comparison screenshots to a pull request using browser automation and GitHub CDN. This captures the UI on the base branch (before) and the PR branch (after), then presents them in a side-by-side comparison table.

Arguments

$ARGUMENTS - PR number and optional description of what to screenshot (e.g. pages/routes to capture)

Execution Steps

1. Parse Arguments & Get Repo Info

# Get current repo info
REPO=$(gh repo view --json nameWithOwner -q .nameWithOwner)

# Get PR info (base branch, head branch, title, body)
gh pr view <PR_NUMBER> --json baseRefName,headRefName,title,body

# Save branch names
BASE_BRANCH=<baseRefName>   # e.g. main
HEAD_BRANCH=<headRefName>   # e.g. feature/new-ui

2. Capture "Before" Screenshots (Base Branch)

# Stash any uncommitted changes, then checkout the base branch
git stash
git checkout $BASE_BRANCH

# Install dependencies (in case they differ)
pnpm install

# Start dev server
pnpm dev &
DEV_PID=$!
sleep 5

# Create screenshot directories
mkdir -p docs/screenshot/before docs/screenshot/after

# Open the page(s) to screenshot
agent-browser open "http://localhost:3000/path/to/feature"
sleep 2

# Login if needed (check CLAUDE.md for credentials)
agent-browser snapshot -i
agent-browser fill @e1 "email"
agent-browser fill @e2 "password"
agent-browser click @e3

# Navigate to the relevant page and take screenshot(s)
agent-browser screenshot docs/screenshot/before/feature-overview.png
agent-browser screenshot docs/screenshot/before/feature-detail.png

# Stop dev server and close browser
agent-browser close
kill $DEV_PID 2>/dev/null || true
pkill -f "next dev" || true

3. Capture "After" Screenshots (PR Branch)

# Checkout the PR branch
git checkout $HEAD_BRANCH
git stash pop || true

# Install dependencies
pnpm install

# Start dev server
pnpm dev &
DEV_PID=$!
sleep 5

# Open the same page(s)
agent-browser open "http://localhost:3000/path/to/feature"
sleep 2

# Login if needed (same steps as before)
agent-browser snapshot -i
agent-browser fill @e1 "email"
agent-browser fill @e2 "password"
agent-browser click @e3

# Navigate to the same page and take screenshot(s)
agent-browser screenshot docs/screenshot/after/feature-overview.png
agent-browser screenshot docs/screenshot/after/feature-detail.png

# Stop dev server and close browser
agent-browser close
kill $DEV_PID 2>/dev/null || true
pkill -f "next dev" || true

4. Screenshot Naming Convention

Use matching filenames in before/ and after/ so the comparison table is easy to construct:

docs/screenshot/
├── before/
│   ├── feature-overview.png
│   ├── feature-detail.png
│   └── feature-empty.png
└── after/
    ├── feature-overview.png
    ├── feature-detail.png
    └── feature-empty.png

Naming ideas:

feature-overview.png — main view
feature-detail.png — expanded/detailed view
feature-empty.png — empty state
feature-loading.png — loading state
feature-mobile.png — mobile viewport

5. Commit & Push

git add docs/screenshot/
git commit -m "$(cat <<'EOF'
docs: add before/after screenshots for [feature]

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
EOF
)"
git push

6. Update PR Body with Before/After Comparison Table

IMPORTANT: Use GitHub CDN URL format with ?raw=true:

https://github.com/<REPO>/blob/<BRANCH>/docs/screenshot/<before|after>/<filename>.png?raw=true

Where <REPO> is the nameWithOwner from step 1 and <BRANCH> is the PR head branch.

Update PR body with a comparison table:

gh pr edit <PR_NUMBER> --body "$(cat <<'EOF'
# PR Title

## Summary
[existing summary]

## Screenshots

### Feature Overview
| Before | After |
|--------|-------|
| ![Before](https://github.com/<REPO>/blob/<BRANCH>/docs/screenshot/before/feature-overview.png?raw=true) | ![After](https://github.com/<REPO>/blob/<BRANCH>/docs/screenshot/after/feature-overview.png?raw=true) |

### Detail View
| Before | After |
|--------|-------|
| ![Before](https://github.com/<REPO>/blob/<BRANCH>/docs/screenshot/before/feature-detail.png?raw=true) | ![After](https://github.com/<REPO>/blob/<BRANCH>/docs/screenshot/after/feature-detail.png?raw=true) |

[rest of PR body]
EOF
)"

Add one table per page/state you screenshot. Each table has a single row with the before image on the left and the after image on the right.

7. Clean Up

agent-browser close
pkill -f "next dev" || true
git checkout $HEAD_BRANCH

GitHub CDN URL Format

Always use this exact format for images to render in GitHub:

https://github.com/OWNER/REPO/blob/BRANCH/path/to/image.png?raw=true

OWNER/REPO: Get from gh repo view --json nameWithOwner -q .nameWithOwner
BRANCH: The PR head branch name (get from gh pr view)
?raw=true: Required suffix for images to display

Authentication

If the app requires login, check the project's CLAUDE.md for test credentials.

Tips

Consistency is key: screenshot the exact same pages/viewports/states on both branches so the comparison is meaningful
Use agent-browser tab list to see all open tabs
Use agent-browser tab N to switch tabs
For elements not in snapshot, use CSS selectors: agent-browser click "button.submit"
Scroll with: agent-browser scroll down 500
For responsive comparisons, resize the viewport: agent-browser eval "window.resizeTo(375, 812)"
If a page doesn't exist on the base branch (new page), skip the "before" screenshot and note it in the PR body as New Page

pr-screenshot