Inspiration

Chat Council started with a question: how can a small Twitch audience make a visible difference to what happens on screen? I compared five gameplay and visual directions with Codex, then chose a short arcade contest where chat becomes the streamer's opponent.

What it does

Chaos Arcade is a five-round game for Twitch streamers and viewers. The host opens a 12-second vote. Viewers type !tiny, !reverse, or !speed to choose the obstacle. The streamer then has eight seconds to stop a moving marker inside the target. Tiny Target narrows the margin for error, Reverse changes direction, and Double Speed makes timing harder. Each result awards a point to the streamer or chat.

One viewer can influence the challenge. Each viewer gets one ballot per voting window, and their latest valid choice replaces the earlier vote. Votes, the selected modifier, and the result stay visible. A clearly labeled rehearsal mode works before connecting Twitch.

How I built it with Codex

I used OpenAI Codex to explore five directions, implement the game state and visual interface, connect Twitch authorization and EventSub, and develop checks for voting and session behavior.

The game uses JavaScript, HTML, CSS, Twitch OAuth, and an EventSub WebSocket. It runs as a static browser app on Vercel. Twitch access is read-only: the game does not send chat messages, require a client secret or second bot account, or make paid LLM calls at runtime. Acid yellow, purple, pink, large controls, and short instructions keep the game legible on screen.

Challenges and what I learned

Live integration exposed a browser-specific fetch binding error and a timing boundary between Twitch's clock and the browser's clock. Codex helped reproduce both, correct the receiver binding, align event timestamps, and add regression tests. Testing the actual authorization-to-chat path mattered.

Voting also needs precise rules: duplicate messages must not count twice, revotes replace earlier choices, late votes must not leak into another round, and connection loss must not silently introduce simulated voters.

What works today

Twenty-two deterministic tests and the production build pass. Browser checks covered a full match, restart, timeouts, disconnected live mode, narrow screens, and a 720p presentation. The corrected 1080p demo shows actual Twitch commands in team_vince's chat: !reverse replaces a prior !tiny vote while the voter count stays at one, then !speed selects Double Speed in the next round. Both rounds show the resulting score; the second also demonstrates the timeout rule. The channel was offline during these tests; this is a hackathon prototype, with no claim of streamer adoption or an active broadcast.

What's next

Test pacing with streamers and viewers, tune voting windows for stream delay, and add new challenges if playtests show the core loop is fun.

Built With

Share this project:

Updates

Submission history