Most AI-generated art is static and forgettable. Anthropic's algorithmic-art skill takes a different approach. It teaches the model to write a manifesto for a generative aesthetic movement first, then express that philosophy through p5.js sketches with seeded randomness. The result is art that lives, different on every refresh and reproducible from the seed.
◇ live · flow field, seeded · click ↻ regen for a new composition
What you are looking at is one composition out of an infinite family. The seed in the corner is the only input, the same seed always produces the same image, and every other run lives somewhere else in the same family. That property, reproducible from a seed, is most of what makes a generative sketch a usable artifact rather than a one-off render.
The interesting part is the prompt-engineering decision underneath, not the rendering output. Forcing the philosophy step before the code step keeps the model from defaulting to template patterns. The output is not "a generative art piece"; it is this specific aesthetic expressed through these specific algorithms. The Vibgineer landing page hero is built on the same idea.
§01What it does
The skill runs in two stages. The first produces a .md philosophy document, a written manifesto for what the piece is supposed to feel like. The second produces the .html and .js sketch that expresses that philosophy. Every output is reproducible because randomness is seeded, so the same seed always produces the same composition. A handful of named parameters (density, speed, color drift) are exposed for tuning, and the whole thing ships as a self-contained HTML viewer that drops into a browser and runs.
§02Why it punches above its weight
The skill is mostly prompting discipline. The actual code is just p5.js, which the model already knows. What changes is the commit. The philosophy step forces specificity, and that specificity becomes the difference between "a flow field" and "this flow field, which moves like wind across a wheat field at dusk." That is the lesson worth porting to other domains: a written commitment to a specific aesthetic before any code gets generated tends to produce work the model could not produce on a default prompt.
§03Setup
# install via the official skills plugin
/plugin install algorithmic-art@anthropic-agent-skills
# usage in claude code:
/skill algorithmic-art "controlled chaos, dusk palette"
The skill takes a brief direction and produces both the philosophy doc and the runnable sketch. Output usually lands in a few seconds; iteration happens by tweaking parameters in the generated HTML.
◆ pull quote
“Algorithmic philosophies are computational aesthetic movements expressed through code. Beauty lives in the process, not the final frame.”
§04Caveats
- p5.js only. For WebGL, three.js, or shaders, this is not the right skill, though the philosophy step generalizes to any creative-coding stack worth using.
- Templates constrain. The skill includes a fixed HTML template with Anthropic branding, which you will want to strip for production use.
- Token usage is real. Two generation steps mean two LLM calls, which is worth it for the quality but is not free.