AI QA Testing Agent

HOW IT WORKS

This AI agent uses Claude to autonomously explore your website like a real user would. It navigates pages, clicks buttons, fills forms, and identifies bugs, UX issues, and broken functionality. Multiple agents explore different user flows in parallel (login, signup, checkout, etc.), branching when they discover decision points.

FOCUSED TESTING

For more targeted testing, paste your recent PR description or a summary of changes into the Testing Goal field. For example: "Test the new checkout flow - added coupon code validation and guest checkout option". This helps the agent focus on the areas most likely to have issues.

CREDENTIALS

The agent may encounter login pages, payment forms, or other authenticated areas. You can provide credentials in two ways:

• Upfront (recommended): Enter credentials one per line (e.g., "Username: myuser" or "PASSWORD=secret")
• On-demand: The agent will pause and ask when it needs credentials you haven't provided and it can't continue without them

Flexible format: Use "Key: value" or "KEY=value". Keys with spaces are auto-normalized (e.g., "Browser OAuth Username" becomes "BROWSER_OAUTH_USERNAME"). If no credentials are provided, the agent may attempt to create credentials (e.g. by signing up for an account).

ENVIRONMENT

• The recommended way to use this tool is to trigger it on your staging environment, and provide it with proper test credentials. It can be ran on a production server, but...
• Never use production credentials or API keys - the agent will interact with your site and may trigger real actions

API USAGE WARNING

This tool uses the Claude API, and token usage is high. A typical session will use millions of tokens.

• Use the Max Duration or Max Cost settings to limit the usage.
• Max cost can be reached quickly. It is recommended to only run this tool prior to deployment, not for standard PRs.