Description
When an agent tests a local web app it often guesses selectors and flails when they miss. Anthropic's Webapp Testing skill gives it a method: write native Python Playwright scripts and look at the rendered page before deciding what to click.
Static HTML is read directly for selectors. Dynamic apps get their servers started with the bundled with_server.py, then the agent loads the page, waits for network idle, screenshots or inspects the DOM, and only then acts. It also says to run scripts with --help first rather than flooding the context with source.
Decision tree: Separates static pages from dynamic apps.
with_server.py: Manages the lifecycle of one or more servers.
Recon then action: Wait for load, inspect, then script.
Screenshots and logs: Capture browser screenshots and read console logs.
Static HTML is read directly for selectors. Dynamic apps get their servers started with the bundled with_server.py, then the agent loads the page, waits for network idle, screenshots or inspects the DOM, and only then acts. It also says to run scripts with --help first rather than flooding the context with source.
Features
Decision tree: Separates static pages from dynamic apps.
with_server.py: Manages the lifecycle of one or more servers.
Recon then action: Wait for load, inspect, then script.
Screenshots and logs: Capture browser screenshots and read console logs.
