Micro Demoscene Engine
shippedA single self-contained HTML file that procedurally generates a synced audiovisual demoscene demo (WebGL shader visuals + Web Audio synthesis) with zero external assets, models, or network calls.
Approach
One index.html with inline CSS/JS, no build step, because shaders are plain strings and the audio engine is a few dozen lines of Web Audio nodes — a bundler would add risk for zero benefit here. A WebGL2 context renders a fullscreen triangle whose fragment shader is swapped on a timeline; 4 shaders (plasma, raymarched tunnel, starfield warp, kaleidoscope) each take uniforms for time, a bass-envelope value, and a palette index, so they visibly react to the beat. A shared `clock` object (bar/beat/bpm/sceneIndex) advances via requestAnimationFrame regardless of audio state, so the visual sequencer auto-advances through scenes on load even before sound is enabled. The soundtrack is a 32-step procedural sequencer (kick/bass/arp/pad built from OscillatorNode + GainNode envelopes, scheduled with the standard lookahead-timer pattern) that writes its current bass envelope value into the shared clock every tick, which is what actually drives the visual reaction — so the sync is real, not simulated. Because browsers refuse to autoplay AudioContext without a user gesture, a single '▶ enable sound' button resumes the AudioContext on click; this is a platform restriction common to every browser demo, not a shortcut, and the visuals are already fully alive before that click.
The source post
https://x.com/AIWorkflowX/status/2118079843271372912
Scoring
Pick
| surprise | 4 |
|---|---|
| demonstrability | 5 |
| self_containedness | 5 |
| honesty | 5 |
Every other build this month has been a dashboard, analyzer, or chat-adjacent tool; this is the first pure-craft demoscene piece, and it lands on every rubric axis cleanly. It's a single HTML file with no dependencies, no accounts, no API keys, and nothing fetched at runtime or build time, since the visuals and audio are procedurally generated in-browser (WebGL shaders + Web Audio API synthesis) rather than pulled from anywhere. A stranger gets it instantly: open the page, see synced motion and sound within seconds, no explanation needed. And the claim is fully testable in the artifact itself, not mocked, since the graphics and audio are genuinely computed live rather than pre-rendered or faked. The only other candidate, Opera's WebAI preview, isn't really a buildable idea at all; it's a vendor announcement about a proposed browser standard gated behind specific Opera builds and devices, which fails both the single-deployable-web-artifact rule and the self-containedness bar.
Review
| shipped | 2 |
|---|---|
| honest | 5 |
| worth_it | 4 |
| efficient | 5 |
No change: the four editable levers (scout query, pick rubric, build lessons, routing executors) all behaved correctly tonight — scout surfaced a strong candidate, pick's reasoning holds up against the actual artifact, build matched the plan with no deviation per NOTES.md, and the codex build executor's 100%-over-42-runs streak continued. The one real problem tonight is that the published URL is dead, and that's a post-publish infra/DNS issue outside scout/pick/build/routing, not a repeatable prompt or executor failure I can address with one of these four levers.
escalated: Not a contract dispute, but worth a human look: the gate only validates the local build output before publish, so nothing in the pipeline would have caught that tonight's published site is unreachable. https://micro-demoscene-engine.fantasia.build has returned Cloudflare 522 on every check for the 7 hours since publish, and the recorded pages.dev deploy URL doesn't resolve at all, while other recent fantasia.build deployments are fine. Worth checking the Cloudflare Pages project for micro-demoscene-engine directly, and separately considering whether gate should include a post-publish live-URL check.
Cost
| total | $0.1228 |
|---|---|
| xai | $0.1228 |
What it looked at
Opera WebAI early preview** (post by @futurepedia_io): Proposed web standard letting sites/extensions run AI models locally in the browser with no API keys or cloud calls; interesting as a standalone, opt-in browser capability runnable in a simple web page on supported Windows/Mac devices.
Tiny HTML audiovisual demoscene generator** (post by @AIWorkflowX): AI autonomously produces a complete, fully functional demoscene (graphics + audio) inside one tiny self-contained HTML file; interesting because the output is a pure browser-renderable artifact with zero dependencies or external models.
Gate
| pass | project directory exists — /Users/artax/code/builds/2026-11-06/project |
|---|---|
| pass | no build step needed — static project |
| pass | build output with index.html — /Users/artax/code/builds/2026-11-06/project |
| pass | index.html is a document — 14928 bytes |
| pass | local asset references resolve |
| pass | page loads without console errors |
Stages
| scout | grok · ok · 41.1s |
|---|---|
| pick | claude · ok · 26.6s |
| plan | claude · ok · 52.9s |
| build | codex · ok · 253.4s |
| gate | local · ok · 2.9s |
| publish | local · ok · 47.5s |
| review | claude · ok · 155.6s |
Notes
Omitted for lack of
None. All visuals and sound are synthesized locally using browser APIs; no service or generated image is needed.