SetupSet up with your AI
Hand chekd to your AI
If you already talk to an AI that can run commands on your machine — Claude Code, Claude Desktop, Codex, Antigravity — you don’t have to set chekd up by hand. Copy the prompt below, paste it in, approve a couple of prompts, and it’s chekingchecking your app.
agent prompt
You are helping me run chekd on my web app. chekd is a local-first verification tool: it opens a real browser, walks a running web app, and reports the dead, broken, and visually-off spots as reviewable cards. Nothing is uploaded to a chekd server.
Do this:
1. Make sure my app is running (a dev server or a deployed URL), then run:
npx chekd@latest <MY_APP_URL>
Replace <MY_APP_URL> with where my app is served, e.g. http://localhost:5173 or https://myapp.com (type the address bare, no angle brackets).
2. It runs on Windows, macOS, and Linux via Node (Node 18+). Use my normal shell (PowerShell, Terminal, or bash).
3. Expect about two approval prompts: one to let npx download the chekd package, and one the first time it downloads its browser (~100MB, one-time). Approving them is normal.
4. When it finishes it prints a local http://127.0.0.1:<port>/ cockpit URL and keeps running. Open that URL so I can review the findings, then stop it with Ctrl+C when I am done.
If it fails:
- "node" or "npx" not found -> Node is not installed. Install Node 18+ from nodejs.org, reopen the shell, and re-run.
- a browser / Chromium / Playwright launch error -> run: npx playwright install chromium then re-run the chekd command.
Report the findings back to me as a short list (what is dead or broken, and where). Do not fix anything until I say so.
What your AI will do
-
Run
npx chekd@latest <your-app-url>Point it at a running dev server or a deployed URL.@latestjust pins the newest published build. -
Clear about two approvals
One to let
npxfetch the package, one the first time it downloads its browser (~100MB, one-time). That is the whole ceremony. -
Open the cockpit and review
chekd prints a local
http://127.0.0.1:<port>/URL and keeps running. Open it, review the findings, then stop it with Ctrl+C.
If something snags
- “node” / “npx” not found — Node isn’t installed. Get Node 18+ from nodejs.org, reopen the shell, and re-run.
- Browser / Chromium / Playwright launch error — run
npx playwright install chromiumonce, then re-run the chekd command. (First run normally auto-installs it; this is only the fallback.)
chekd runs locally. A browser-tab-only assistant (like chatgpt.com) can’t install or launch a local tool — so this on-ramp is for an AI that runs on your machine. That’s the honest boundary today.
Local records · explicit model use
chekd stores findings, corrections, and screenshots on your machine and does not upload them to a chekd service. The objective checks are model-free and cost $0; any model-backed step is BYO-model and explicit.