Pre-Post Screenshot Skill
Package:
pre-postVisual diff tool for PRs — fastest path from code change to visual documentation.
Agent Behavior Rules
DO NOT:
- •Switch git branches, stash changes, start dev servers, or assume what "before" is
- •Use
--fullunless user explicitly asks for full page / full scroll capture - •Post screenshots to PR without user approval
DO:
- •Use
--markdownwhen user wants PR integration or markdown output - •Use
--responsiveto capture both desktop and mobile viewports - •Use
--mobile/--tabletif user mentions phone, mobile, tablet, responsive - •Assume current state is After (localhost = after, production = before)
- •Show screenshots to user before posting to PR
- •If user provides only one URL, ASK: "What URL should I use for the 'before' state? (production URL, preview deployment, or another local port)"
Execution Order
1. Pre-flight Checks
# Detect running dev server lsof -i :3000 2>/dev/null || lsof -i :3001 2>/dev/null || lsof -i :5173 2>/dev/null || lsof -i :8080 2>/dev/null
If no dev server is running, tell the user to start one.
# Check production URL is accessible
curl -s -o /dev/null -w "%{http_code}" "<production-url>"
- •200 → proceed
- •401/403 → warn user: "Production URL requires authentication. Options: (1) provide a public URL, (2) skip 'before' and capture after-only, (3) provide auth cookies"
- •No production URL → "after-only" mode: screenshot localhost only, label as current state
2. Route Detection + Refinement
# Detect affected routes from git diff npx pre-post detect
This outputs JSON with detected routes, confidence levels, and source files.
Claude's role: Review the JSON output using conversation context:
- •Add routes you know are affected from the work done in this session
- •Remove false positives (e.g., API-only changes)
- •For dynamic routes (e.g.,
/blog/[slug]), ask user for a sample value - •Present to user: "I'll screenshot these routes:
/dashboard,/settings. Want to add or change any?"
3. Screenshot Capture
Option A: CLI (preferred — deterministic)
# Single route, desktop only npx pre-post compare \ --before-base https://prod.com \ --after-base http://localhost:3000 \ --routes /dashboard \ --output /tmp/pre-post # Multiple routes, responsive (desktop + mobile) npx pre-post compare \ --before-base https://prod.com \ --after-base http://localhost:3000 \ --routes /dashboard,/settings,/ \ --responsive \ --output /tmp/pre-post
Option B: Playwright MCP (for more control)
Use when you need custom waits, interactions, or complex page states:
browser_resize(1280, 800)
browser_navigate("https://prod.com/dashboard")
browser_wait_for(time: 3)
browser_take_screenshot(filename: "/tmp/pre-post/dashboard-desktop-before.png")
browser_navigate("http://localhost:3000/dashboard")
browser_wait_for(time: 3)
browser_take_screenshot(filename: "/tmp/pre-post/dashboard-desktop-after.png")
# Mobile
browser_resize(375, 812)
browser_navigate("https://prod.com/dashboard")
browser_wait_for(time: 3)
browser_take_screenshot(filename: "/tmp/pre-post/dashboard-mobile-before.png")
browser_navigate("http://localhost:3000/dashboard")
browser_wait_for(time: 3)
browser_take_screenshot(filename: "/tmp/pre-post/dashboard-mobile-after.png")
4. User Approval
Show screenshots in conversation. Ask: "Here are the before/after screenshots. Should I post to PR, retake any, or add more pages?"
5. Upload + PR Markdown
# Upload and generate markdown mkdir -p /tmp/pre-post ./scripts/upload-and-copy.sh /tmp/pre-post/before.png /tmp/pre-post/after.png --markdown
Or use the CLI's built-in upload:
npx pre-post <before.png> <after.png> --markdown
For multi-route PRs, generate this format:
## Visual Changes ### `/dashboard` <details open> <summary>Desktop (1280x800)</summary> | Pre | Post | |:---:|:----:| |  |  | </details> <details> <summary>Mobile (375x812)</summary> | Pre | Post | |:---:|:----:| |  |  | </details> --- *Captured by [pre-post](https://github.com/juangadm/pre-post)*
6. PR Integration
# Get current PR gh pr view --json number,body # Append screenshots to PR body gh pr edit <number> --body "<existing-body> <generated-markdown>"
If no gh CLI: output markdown and tell user to paste manually.
Quick Reference
# Basic usage (two URLs) pre-post site.com localhost:3000 # Detect routes from git diff pre-post detect pre-post detect --framework nextjs-app # Compare with auto-detected routes pre-post run --before-base https://prod.com --after-base http://localhost:3000 # Compare specific routes pre-post compare --before-base URL --after-base URL --routes /dashboard,/settings # Responsive (desktop + mobile) pre-post compare --before-base URL --after-base URL --responsive # From existing images pre-post before.png after.png --markdown # Via npx npx pre-post detect npx pre-post compare --before-base URL --after-base URL
| Flag | Description |
|---|---|
-m, --mobile | Mobile viewport (375x812) |
-t, --tablet | Tablet viewport (768x1024) |
--size <WxH> | Custom viewport |
-f, --full | Full scrollable page |
-s, --selector | CSS selector to capture |
-r, --responsive | Desktop + mobile capture |
--routes <paths> | Explicit route list (comma-separated) |
--max-routes <n> | Max detected routes (default: 5) |
--framework <name> | Force framework detection |
--before-base <url> | Production URL |
--after-base <url> | Localhost URL |
-o, --output | Output directory (default: ~/Downloads) |
--markdown | Upload images & output markdown |
--upload-url <url> | Upload endpoint (overrides git-native default) |
Image Upload
Screenshots are committed to .pre-post/ on the current PR branch and served via
GitHub blob URLs pinned to the commit SHA. This works for both public and private
repos — blob URLs are same-origin on GitHub, so the markdown renderer resolves them
with the viewer's authentication.
# Default (git-native — commits to PR branch, works on any repo) ./scripts/upload-and-copy.sh before.png after.png --markdown # Explicit override for external storage: IMAGE_ADAPTER=0x0st ./scripts/upload-and-copy.sh before.png after.png --markdown
Error Reference
| Error | Fix |
|---|---|
command not found | npm install -g @juangadm/pre-post |
browserType.launch: Executable doesn't exist | npx playwright install chromium |
| 401/403 on production URL | See pre-flight section above |
| Element not found | Verify selector exists on page |
| No changed files detected | Specify routes manually with --routes |
| Could not determine commit SHA | Ensure git push succeeded and HEAD is valid |