From verdict
Token-efficient browser verification for AI coding agents. Verifies pages, clicks, forms, CSS, screenshots. Use for browser testing instead of Playwright MCP.
How this skill is triggered — by the user, by Claude, or both
Slash command
/verdict:verdictThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Token-efficient browser verification via Bash. **Use this instead of Playwright MCP for all browser verification.**
Token-efficient browser verification via Bash. Use this instead of Playwright MCP for all browser verification.
cd <project-root> && npm install verdict-cli 2>/dev/null || true
B="node_modules/.bin/verdict"
# Navigate and verify
node $B goto <url>
node $B snapshot -i # ARIA refs for interactive elements
node $B click @e3 # Click by ref
node $B fill @e5 "text" # Fill input
node $B snapshot -D # Diff: verify action changed page
# Inspect
node $B css @e3 font-size # Computed CSS value
node $B inspect @e3 # Full box model + styles
node $B console # JS errors
node $B network # Network requests
# Visual
node $B screenshot /tmp/page.png
node $B responsive /tmp # Mobile + tablet + desktop
node $B viewport 375x812 # Set viewport
# Auth (for pages requiring login)
node $B handoff # Opens visible Chrome for manual login
# ... log in ...
node $B resume # Back to headless
node $B auth-save myapp # Save session encrypted
node $B goto-auth <url> --profile myapp # Auto-load auth
| I want to... | Command |
|---|---|
| Check page loads | goto <url> then snapshot -i |
| Verify action changed page | snapshot -D |
| Check CSS value | css @eN <property> |
| Full box model + styles | inspect @eN |
| Take screenshot | screenshot /tmp/page.png |
| Test form | fill @eN "val" then click @eM then snapshot -D |
| Check JS errors | console |
| Test responsive | responsive /tmp |
| Multiple checks at once | chain [["goto","url"],["snapshot","-i"],["console"]] |
QA Tool: PLAYWRIGHT_MCPnpx claudepluginhub tuandm/verdict-cliVerifies browser-rendered HTML, UI, visual docs, presentations, local apps, and browser-facing changes in a real browser. Checks layout, responsiveness, overflow, console errors, and network issues.
Provides browser-level verification using Claude-in-Chrome (primary) or chrome-devtools-mcp (fallback). Includes provider detection, dev server detection, and graceful degradation.
Automates browser tasks like E2E testing, form filling, screenshots, and scraping using Vercel's agent-browser CLI with ref-based element targeting.