What problem does it solve? Building web games without automated verification leads to broken controls, invisible elements, and console errors that only surface during manual play. This Skill enforces a disciplined implement-act-pause-observe-adjust loop so every change to an HTML/JS game is exercised with real inputs and visually confirmed before moving on. ## Core Features & Use Cases - Playwright action loop: Runs scripted keyboard and mouse bursts against the game via a dedicated client script, capturing screenshots, render_game_to_text state, and console errors after each iteration. - Deterministic time stepping: Uses a window.advanceTime(ms) hook so frames advance predictably during automated tests instead of relying on flaky real-time loops. - Structured state output: Requires a window.render_game_to_text function returning concise JSON of player, entities, score, and mode so state can be verified without visuals. - Progress handoff: Maintains a progress.md file with the original prompt, TODOs, and notes so another agent can continue the work seamlessly. - Use Case: While adding a shooting mechanic to a canvas game, run the client script with an action payload, inspect the latest screenshot to confirm the enemy disappears at zero health, verify the score updates in the text state, and fix the first console error before iterating. ## Quick Start Ask the agent to build or modify the web game and run the Playwright test script with an action payload, then inspect the screenshots and render_game_to_text output to confirm the change works.
