I Built a Full Game in One Afternoon (GLM 5.3 + Supabase)

Resources: Supabase: https://supabase.plug.dev/vhnyrIK Supabase Agent Skills: https://github.com/supabase/agent-skills Supabase MCP: https://supabase.com/docs/guides/ai-tools/mcp Railway: https://railway.com GLM 5.3 release: https://z.ai/blog/glm-5.3 GLM 5.3 shipped a complete game — with accounts and deploy — from one prompt, in one afternoon. Full tutorial: the prompt, the test harness, Railway, and Supabase auth. In this GLM 5.3 tutorial I test Z.ai's newest frontier open-weights model the day after release, inside the ZCode IDE. One spec-style prompt produces a complete Angry Birds-style physics game — Matter.js, five levels, scoring, star ratings — in a single HTML file. Then the workflow that actually matters: iterating with short prompts (four new bird types, TNT, a level-select screen), verifying every change with a headless test harness that plays the whole game with the real physics engine, deploying to Railway in two commands, and adding real user authentication with Supabase — GitHub OAuth plus email login codes, with row-level security protecting each player's saves. The Supabase agent skills and MCP server are installed live, entirely through prompting. Every number I quote about GLM 5.3 comes from Z.ai's release post, and the full source code is free below — including PROMPT.md, the exact prompt that started the build. The game was never the point: this is the pattern for shipping production apps that other people can use. TIMESTAMPS: 0:00 - The shipped game (GLM 5.3 demo) 0:34 - Demos aren't products: the real problem 1:05 - GLM 5.3 fly-by: benchmarks and open weights 2:08 - The one prompt (spec, not vibes) 3:20 - First result: one HTML file, playable 3:49 - Iterate, don't regenerate 4:41 - The test harness: making the AI prove it 5:11 - Railway agent skill, installed by prompt 5:47 - Deploy to Railway 6:27 - Supabase agent skills + MCP 7:16 - User authentication: GitHub + email codes 8:14 - Free source code giveaway 8:38 - The pattern: spec, build, prove, ship, secure 🎁 Free source code (MIT): The complete build — game, headless test harness, deploy config, auth setup, and PROMPT.md with the exact starter prompt 👉 https://github.com/derekcheungsa/slingshot-siege ▶ Play it live: https://slingshot-siege-production.up.railway.app 🎓 Hermes Deep Dive Course (Udemy): Master building autonomous AI workflows with Hermes Agent 👉 https://www.udemy.com/course/hermes-agent-masterclass-build-autonomous-ai-workflows/?couponCode=HERMES_AGENT 📧 FREE Weekly AI Automation Newsletter: Get weekly tutorials, free workflow templates, and the latest AI news straight to your inbox 👉 https://aifornoncoders.com/newsletter 🤝 Join the AI Automation Engineering Community: Go deeper — interact with me and the community, get feedback on your builds, and stay ahead of the curve 👉 https://www.skool.com/ai-automation-engineering-3014 If you fork the repo and ship your own version, drop a comment with the link — I feature the best builds. #GLM53 #ZCode #Supabase #AITutorial #VibeCoding