⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
genart
>
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
genart — a generative art skill for Claude Code
Create long-form generative art with AI assistance — deterministic, hash-seeded, ready for onchain platforms like Art Blocks, 256ART, Verse, Highlight, Plottables and bootloader.art — or for screens, via Artpoint.
This Claude Code plugin teaches Claude the working knowledge of the field: seeding a PRNG from a token hash, rendering the same composition at any resolution, designing traits and rarity tables, signalling preview capture, exporting for print and pen plotters, and where the ethical lines of generative art are. It covers creative coding with Canvas 2D, p5.js, three.js/WebGL and SVG — and ships runnable tools to verify that a sketch really is deterministic before you mint it.
Made by a generative artist, for generative artists using AI as a studio assistant — not as the artist.

Six seeds of Impossible Machine (Camille Roux) — hand-drawn technical blueprints, one hash each. The edition-level view the scripts are there to give you: same code, same rules, one sheet per seed.
Install
/plugin marketplace add camilleroux/genart-skill
/plugin install genart@camilleroux-genart
Then just talk about your sketch — the skill loads on its own. Or invoke it with
/genart.
What's in it
One skill, genart, with reference sheets loaded on demand:
- determinism — PRNG choice, seeding from a hash, named sub-streams, the things that silently break reproducibility
- resolution — rendering the same piece at 400px and at 4000px
- features — designing traits and rarity tables that survive an edition
- ethics — originality, attribution, licences, AI disclosure
- tooling — debug GUI, keyboard shortcuts, PNG and video export
- verification — checking a sketch before minting, and what a test can't prove
- platforms — Art Blocks, ABX, 256ART, Verse, Highlight, Plottables, bootloader.art, self-hosted, Artpoint (art on screen)
Checking and rendering
The plugin ships no code — only Markdown. When a sketch needs checking or
rendering, Claude writes a small Playwright harness in your project,
following the spec in references/verification.md, and reuses it from then on:
- check — repeatability, distinctness, global-state contamination (A, B, A in one page), feature stability
- one render — a PNG Claude looks at before judging any visual change
- contact sheet, census (rarity, measured), batch (full-size PNGs)
The harness lives in your repo, so you can read it, run it yourself, and keep it
in CI. It needs Playwright as a dev dependency (npm i -D playwright && npx playwright install chromium). Claude proves it once by making it fail on a
copy of your sketch with a Math.random() left in.
Why no shipped scripts: Anthropic's plugin directory accepts code only from Verified organizations, and the checks are short enough to rebuild from a precise spec.
The contract a sketch must follow (three lines of shim for any platform) and
what the checks can and cannot prove are in references/verification.md.
How the platform sheets work
They contain no version numbers, no figures and no field names — only the stable mental model of each platform, the canonical doc URLs, and the questions to ask those docs. Claude fetches the real documentation before writing platform code.
That is deliberate. A copied value goes stale in months and is then worse than nothing, because you believe it. A link stays right.
On determinism claims
Nothing here can prove cross-machine determinism, and this plugin does not claim
to. In WebGL, shader compilers, float precision, rasterisers and MSAA differ
between GPUs; in JS, transcendental functions differ between engines. What is
actually checkable is same-machine reproducibility, perceptual stability across
sizes, and feature stability. references/verification.md is explicit about
where the line falls.
Maintenance
CI checks monthly that the URLs in the sheets still resolve — confirmed 404/DNS only, bot walls don't count — opening an issue when one dies. The sheets contain no volatile facts, so dead links are the only thing that rots.
Show what you make
If you mint something with it, I would genuinely like to see it — open a Discussion with a render and the platform you released on, or post it and tag @camillerouxart. Pieces made with the skill get shared onward.
Reference sheets are the easiest thing to contribute: a platform that is missing, a doc URL that moved, a mental model that is wrong. Issues and PRs welcome — the editorial rules are short.
If it saved you an afternoon, a ⭐ is how the next artist finds it.
Author
Built and maintained by Camille Roux — generative/algorithmic artist. Maintainer of awesome-generative-art, the curated list of generative art platforms, libraries and resources.
Follow the art: art.camilleroux.com · X · Instagram · Farcaster · Bluesky · Mastodon · TikTok
License
MIT.
// HOW IT'S BUILT
// FILES
2 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
skills/genart/SKILL.md7.6 KBupstream ↗
--- name: genart description: > Craft long-form generative art for onchain platforms — Art Blocks, 256ART, Verse, Highlight, Plottables, bootloader.art, or a self-hosted drop — and for screen diffusion channels such as Artpoint. Covers hash-seeded determinism, resolution-agnostic rendering, features and rarity design, preview capture signals, debug GUIs and image/video export shortcuts, verification of a sketch before minting, and the ethics of the field. Use when the user mentions generative art, gen art, creative coding, long-form, onchain art, a seeded sketch, a PRNG or deterministic randomness, traits, rarity, features, a mint, a plotter or SVG output, art on screen or delivering a piece as a video, or pastes platform APIs such as tokenData, $features, inputData, $bootloader, BTLDR, hl-gen, or a base64 payload query param. user-invokable: true argument-hint: "[platform|check|render] [path]" license: MIT metadata: author: Camille Roux category: generative-art --- # Generative art Help design, write and check long-form generative artworks — pieces where one algorithm produces a whole edition, each output derived from a hash the blockchain hands you at mint time. ## Default practices, and when to break them These are defaults that work most of the time, not rules. Every one of them has legitimate counter-examples, and those are often the interesting pieces. State the default, then state what breaking it costs. | Default | Breaking it is legitimate when… | |---|---| | All variation comes from the hash | The piece deliberately reads live onchain state (owner, block data). It is then no longer reproducible from the hash alone — own that choice and document it | | One seeded PRNG, fixed consumption order | Almost never. But *named sub-streams* let you add a draw without shifting the whole edition — that is how you get flexibility without breaking the property | | No network, no CDN, no system fonts | Some platforms have an explicit external-asset mode, under their own rules | | No wall clock | The piece is about time passing, or is genuinely interactive. The preview render must stay stable regardless | | Sizes relative to the canvas dimension | The piece is deliberately adaptive and reveals more detail with more room. A choice, not an accident — then know what size it will be captured at | | Runs on current browsers and devices — Chrome, Firefox, Safari, mobile included | A piece built on a bleeding-edge API (WebGPU…) or needing real GPU power is a choice: state the requirement, keep the preview/capture path rendering everywhere, and remember collectors open links on phones | | Emit the platform's render-done signal | Never. A blank thumbnail is a failure with no upside | | Features computed before render, from the seed only | Where the platform computes features outside a browser, it is required. Elsewhere it is a convenience | | Debug code is stripped from the submitted build | Where code is stored unminified and readable onchain, formatting is part of the work | | Never claim "it's deterministic" without testing it, and without stating the scope | Never | **The firm part:** breaking a default must be the artist's explicit choice, never an accident. Point out the deviation, explain what it costs, then do what the artist asked. Do not refuse and do not lecture. ## Ethics - Never reproduce a named living artist's signature work. Techniques are shared heritage; a body of work is not. Offered "make me a <artist>", decline that framing and offer the underlying technique instead. - Credit the algorithm, the shader, the palette you borrowed. - Check that a library's licence survives being written onchain forever. - Be straight about what AI did: writing the code and generating the image are different claims. Full version: `references/ethics.md`. ## Platforms: read the docs, every time Platform APIs change and the fiches in `references/platforms/` deliberately hold no versions, no numbers and no field names — only the stable mental model, the canonical URLs and the questions to ask. So the sequence is always: **open the fiche → fetch the URLs it lists → then write code.** Never write platform-specific code from memory, and never from the fiche alone. If a doc is unreachable, say so and flag what could not be confirmed. | Target | Fiche | |---|---| | Art Blocks (incl. Engine / Flex) | `references/platforms/artblocks.md` | | 256ART | `references/platforms/256art.md` | | Verse | `references/platforms/verse.md` | | Highlight | `references/platforms/highlight.md` | | Plottables (pen plotter, AB Engine) | `references/platforms/plottables.md` | | bootloader.art — ask *which* bootloader | `references/platforms/bootloader.md` | | Self-hosted / no platform | `references/platforms/self-hosted.md` | | Artpoint — art on screen, you deliver a video, not code | `references/platforms/artpoint.md` | | Undecided between two | `references/platforms/comparison.md` | ## Where to look | When the question is about… | Read | |---|---| | Seeding from a hash, PRNG choice, sub-streams, distributions, things that silently break reproducibility | `references/determinism.md` | | Output that changes with canvas size, stroke weights, noise frequency, element density, print or plotter output | `references/resolution.md` | | Designing traits, rarity tables, weights, distribution that came out wrong | `references/features.md` | | Originality, attribution, licences, disclosing AI use | `references/ethics.md` | | Debug panel, param tweaking, keyboard shortcuts, PNG and video export, contact sheets | `references/tooling.md` | | "Is my sketch actually deterministic?", pre-mint checking, what a test can and cannot prove | `references/verification.md` | | Techniques, tutorials, libraries, inspiration, what other artists use | Fetch https://github.com/camilleroux/awesome-generative-art — a maintained list; do not paraphrase it from memory | ## Runnable scripts Two scripts ship with the plugin and run **in place** — never copy them into a project. They need Playwright installed in the artist's project and print the install command if it is missing. Contract and details: `references/verification.md`. | Command | Does | |---|---| | `node "$CLAUDE_PLUGIN_ROOT/scripts/check.mjs" <dir>` | Determinism: repeatability, distinctness, global-state contamination, feature stability | | `node "$CLAUDE_PLUGIN_ROOT/scripts/render.mjs" <dir> --hash 0x…` | One PNG — render it, then **look at it** with Read before judging any visual change | | `node "$CLAUDE_PLUGIN_ROOT/scripts/render.mjs" <dir> --grid N` | Contact sheet, hash + features under each tile | | `node "$CLAUDE_PLUGIN_ROOT/scripts/render.mjs" <dir> --census N --edition-size E` | Real feature distribution vs the rarity table | | `node "$CLAUDE_PLUGIN_ROOT/scripts/render.mjs" <dir> --batch N --size 2400` | N individual full-res PNGs, hash-named — portfolio/print export | When asked to change how a piece looks, close the loop yourself: edit → render one hash → Read the PNG → judge → adjust. Do not describe a visual change you have not looked at. ## Workflow 1. **Target first.** Which platform? If undecided, `references/platforms/comparison.md`. Once decided, open its fiche and fetch its docs before writing anything. 2. **Decide the rarity table before coding.** It is much harder to retrofit. 3. **Build.** Defaults above; deviations discussed, not silently taken. 4. **Look at many outputs, not one.** A single good render proves nothing — `render.mjs --grid` exists for this. 5. **Check before minting** — `check.mjs`, then `references/verification.md` for what the green result does and does not prove.README.mdupstream ↗
Contents are not inlined in the catalog. Open the upstream file.
Plain-text copy, no login: /skill-bundles/camilleroux-genart-skill-skills-genart.md