Abhilash Menon
← Building with AI

AI playtesting for games (playcheck)

Published October 17, 2026

An open-source tool that lets an AI player play a browser game and report where it stops being fun: stuck, confusing, or too hard too soon.

AI coding agents can build a working game in an afternoon. Working is not the same as fun. The tests pass while real players quit in the first 30 seconds.

playcheck tests for that gap. It is an open-source TypeScript tool, published on npm under the MIT licence.

How playcheck tests a game for funPlaywright loads the browser game in Chromium. An AI player gets a screenshot every 2 seconds and sends back the next input. Fun heuristics watch for five problems: stuck, softlock, objective clarity, difficulty cliff and control discoverability. Findings go to an interactive triage, then a Markdown report with severity and a suggested fix.Browser gamePlaywright+ ChromiumAI playerAnthropic, OpenAI or OllamaFun heuristicsStuckSoftlockObjective clarityDifficulty cliffControl discoverabilityInteractive triageMarkdown reportseverity + fixscreenshot every 2 snext input
  1. Playwright launches Chromium and loads the game.
  2. An AI player plays it. Every 2 seconds it takes a screenshot and sends it to a model. Anthropic, OpenAI or a local Ollama model all work.
  3. Heuristics watch for five problems:
    • Stuck: nothing on screen changes for more than 10 seconds.
    • Softlock: 3 or more frames with no way to recover.
    • Objective clarity: no clear goal within 30 seconds.
    • Difficulty cliff: a death in the first 15 seconds, or 3 or more deaths in 30 seconds.
    • Control discoverability: the player cannot work out the controls.
  4. After the run, I triage the findings in an interactive terminal review.
  5. The output is a Markdown report. Each issue has a severity and a suggested fix.

Options let me set the genre, the run length and control hints.

Source on GitHub: github.com/abhilashme/playcheck