QA for AI-coded software

Your AI wrote it.
Did it actually work?

Install the GitHub App. Every pull request gets an AI agent that drives your preview deploy like a human — pass/fail verdict and a recording, posted as a comment before you review.

Try it free →

Free plan · 15 PR checks / month · no credit card

How it works · Use the CLI · GitHub →
usefulagents/storefront #274 Open
feat: redesign checkout with one-click pay
screencli[bot] commented just now · verified preview
✓ All 4 checks passed · preview-pr-274.vercel.app
  • ✓ Checkout flow completes with test card
  • ✓ Success page renders order ID
  • ✓ Confirmation email button is visible
  • ✓ No console errors during flow
▶ Watch recording (38s) →

Built for every AI coding agent

Claude Code Cursor Windsurf Cline Aider OpenCode Antigravity Continue

Install the agent skill: npx skills add usefulagents/screencli --skill screencli

01

One engine.
Three jobs.

A real browser, driven by an AI agent. Use it on every PR, in your dev loop, or to record demos.

✅

Verify every PR

Install the GitHub App. screencli reads each PR diff, runs an AI browser test on your preview deploy, and posts a pass/fail comment with a recording — before you review.

# "Open the checkout flow and complete a test purchase. Confirm the success page renders."
🤖

Check your AI's work locally

Your coding agent shipped a feature. Have it verify the change before you look at it. The CLI runs from inside Claude Code, Cursor, Windsurf — anything that runs shell.

-p "Verify the new settings page saves changes and the toast appears"
🎬

Record polished demos

Same engine, different output. Point it at your app and get a shareable demo video with gradient background, auto-zoom, and click highlights. No editor required.

-p "Walk through the new onboarding flow and show the completed dashboard"
02

The AI drives.
You get the verdict.

Natural language, real-browser testing, and studio-grade output — for PRs, dev loops, and demos alike.

pr-diff

Plans checks from your diff

The GitHub App reads each PR, decides what to test, and runs it against your preview deploy. No test files to write, no selectors to maintain.

verdict

Pass/fail with evidence

Every run produces a structured pass or fail, posted as a PR comment with a recording link. Stop guessing whether flaky CI means the feature actually works.

prompt

Natural language control

Describe what to do in plain English. The AI agent navigates, clicks, types, and scrolls — no scripted sequences or selectors.

auto

Auto post-production

Idle time trimmed. Click positions highlighted. Smooth cursor trails rendered. Auto-zoom to the action. Zero manual editing.

--login

Auth handoff

You log in manually, press Enter, and the AI takes over your authenticated session. Login is never recorded. Sessions are saved for reuse.

--preset

Export everywhere

One command per platform. YouTube, Twitter, Instagram, TikTok, LinkedIn, and GitHub GIF presets built in.

03

Six backgrounds.
Choose your vibe.

Gradient backgrounds with rounded corners and drop shadows. Screen Studio quality, from your terminal.

app.example.com/dashboard
04

The PR flow.
Verdict + recording, in the comment.

Install once. Every pull request gets browser-tested by an AI agent.

01

Install the App

One click. Pick the repos screencli should watch.

github.com/apps/screencli
02

Open a PR

screencli reads the diff, plans browser checks, and runs them against your preview deploy.

✓ Detected preview — pr-274.vercel.app
03

Get the verdict

Pass/fail comment lands on the PR with a recording link. Failures show exactly where it broke.

✓ Comment posted on PR #274
05

The CLI flow.
One command. Shareable link.

For one-off recordings, demos, and AI-agent-driven verification from your local dev loop.

01

Install

Sign in or paste your API key. Optionally install the skill for your coding agent.

$ npx screencli init
02

Record

Tell the agent what to do. It drives the browser, you get a polished video.

$ screencli record https://your-app.com \
    -p "Walk through onboarding"
03

Auto-upload

The recording is composed with effects and uploaded automatically. You get a link.

✓ Uploaded
https://screencli.sh/v/a3f2c8e1
04

Export

Need a specific format? Export for any platform with one command.

$ npx screencli export ./recordings/a3f2c8e1 \
    --preset twitter
06

Simple pricing.

Start free. Upgrade when you need more.

Free
$0
15 credits / month
  • ✓ All effects & gradients
  • ✓ All export presets
  • ✓ Auto-upload & shareable links
  • ✓ Auth handoff
  • — Unlisted recordings
Get started
Pro
Pro
$12
100 credits / month
  • ✓ Everything in Free
  • ✓ Unlisted recordings
  • ✓ Cloud re-rendering
  • ✓ Priority support
  • ✓ Additional credits at $0.10 each
Start Pro

1 credit = 1 PR check or ~10 CLI agent steps. No credit card required for Free.

07

Quick reference.

Everything you need, at a glance.

Record

npx screencli record <url> -p "prompt"

# With gradient background
npx screencli record https://app.com \
  -p "Demo the dashboard" \
  --background aurora

# With authentication
npx screencli record https://app.com \
  -p "Demo features" \
  --login --auth myapp

Export presets

npx screencli export <dir> --preset <name>

# Available presets
youtube    1920x1080  16:9  mp4
twitter    1280x720   16:9  mp4
instagram  1080x1920  9:16  mp4
tiktok     1080x1920  9:16  mp4
linkedin   1080x1080  1:1   mp4
github-gif  800x450   16:9  gif

Cloud commands

npx screencli login
npx screencli logout
npx screencli whoami
npx screencli recordings
npx screencli upload <dir>
npx screencli delete <id>
npx screencli render <id> --background sunset

Backgrounds & effects

--background <name>
  aurora  sunset  ocean
  lavender  mint  ember

--padding <%>        default 8
--corner-radius <px> default 12
--no-shadow  --no-zoom
--no-highlight  --no-cursor
08

Frequently asked

screencli verifies AI-coded changes by driving the browser like a human. Install the GitHub App and every pull request gets an AI-driven browser test on your preview deploy — pass/fail verdict and a recording, posted as a PR comment. Or call the same engine from the CLI inside any AI coding agent (Claude Code, Cursor, Windsurf, etc.) for one-off recordings, demos, and verifications.
Install the screencli GitHub App and pick which repos it watches. When a pull request opens, screencli reads the diff, plans browser checks, runs them against your preview deployment, and posts a pass/fail comment with a recording link — all before you review. You don't write test files; the AI decides what to test based on what changed.
The CLI works standalone. Run npx screencli record <url> -p "your prompt" from any terminal or AI coding agent — no GitHub App required. The App is for teams that want automatic verification on every PR; the CLI is for one-off recordings, demos, and AI-agent-driven verification from your local dev loop.
Your preview deployment. screencli picks up the deployment URL from your GitHub deployments (Vercel, Netlify, Cloudflare Pages, etc.) and runs the AI browser test against it — your production app is never touched.
No install needed — just run npx screencli record <url> -p "your prompt". On first run, the CLI opens a browser for GitHub login, then you're ready to record. For permanent installation, run npm install -g screencli.
Run: npx screencli record https://your-app.com -p "Walk through the main features". The AI agent opens the URL, follows your prompt instructions, and records everything. When it finishes, you get a composed video and a shareable link at screencli.sh. No setup, accounts, or API keys needed for the first run.
Yes. Use the --login flag to log in manually first, then the AI agent takes over your authenticated session. Add --auth myapp to save the session for future recordings: npx screencli record https://app.example.com -p "Show the dashboard" --login --auth myapp. Next time, just pass --auth myapp to reuse the saved session.
Record first, then export with the github-gif preset: npx screencli export ./recordings/your-recording --preset github-gif. This produces an optimized 800x450 GIF (max 12 seconds, under 8 MB) ready to embed in Markdown. Other presets: youtube, twitter, instagram, tiktok, and linkedin.
Any AI coding agent that can run shell commands can use screencli — it's a standard CLI tool. For Claude Code specifically, install the skill with npx skills add usefulagents/screencli --skill screencli, and Claude can record demos autonomously as part of your development workflow.
The CLI is fully free and open source (MIT license) — all video effects, gradient backgrounds, and export presets are included with no limits. The free cloud tier includes 10 recordings per month with shareable links. The Pro plan ($12/month) adds 100 recordings, full analytics, and removes the watermark.
Every recording automatically gets: idle time trimming (removes pauses between actions), auto-zoom (follows the active area), click highlights (visual pulse on clicks), and cursor trails. You also choose from 6 gradient backgrounds (aurora, sunset, ocean, lavender, mint, ember) with customizable padding, corner radius, and drop shadow.
Yes. By default, every recording auto-uploads to screencli.sh and you get a shareable link like screencli.sh/v/a3f2c8e1. The link shows an embedded video player with chapter markers, metadata, and a download button. Use --local to skip the upload and keep recordings local only.