Claude Code has grown beyond simple code generation. After the release of Claude Opus 4.5, developers now face a new bottleneck: verifying that the code a model writes will run as expected. End‑to‑end tests that pump the browser and click through the UI let the agent prove its own work.

The author builds a prompt that instructs the model to implement, verify, and iterate until a full pass. The tool hinges on Playwright MCP—a browser driver that offers more stable interaction than the default Chrome MCP. The prompt also calls for the slash goal command, which forces the model to loop until the goal is met.

For production workloads, the author schedules an Open Claw agent to run a daily sweep of the application. The automated script launches the site, exercises every feature, and reports failures back to the team. This routine surfaces bugs before users encounter them.

The combination of a powerful language model, a high‑fidelity browser driver, and a goal‑driven loop creates a self‑testing pipeline. Engineers can shift from manual QA to focused code reviews, cutting cycle times and reducing regression risk.