agentleFS
ckryptickunalckryptickunal/OpenDesigner/.claude/skills/opendesigner-paint/SKILL.md

opendesigner-paint

Skill56 starsChanged 5 days agoMITView on GitHub

Turns a design system into original painted art made in code (watercolour, pointillism, charcoal, sketch, woodcut, 2D cel): hero films, launch reels, loops and paintings in the brand's palette.

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 0 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run paint.py (scripts/paint.py) from the person's project root.

    line 24
  2. Very specific:

    Determinism is the engine's contract: All randomness goes through the seeded kit, never Math.random.

    line 44
  3. Very specific:

    Read DESIGN.md and opendesigner/state.json:

    line 49
  4. Very specific:

    Without a design system: use --accent '#hex', and offer the opendesigner interview first.

    line 54
  5. Very specific:

    Look at it in the running app, on phone and desktop, light and dark, with reduced motion on and off.

    line 109
  6. Very specific:

    If something was missing, wrong or confusing, record it with engine.py feedback "..." --kind gap|bug|confusing|idea (references/improve.md …

    line 117

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

OpenDesigner: painted art and films

Use this when someone wants original art made by code rather than found art. Typical asks: a painted hero, a launch film or reel, an animated brand moment, a celebration loop, a painted link-preview image, or "something that looks hand-made, in our colours". It is also an answer for three designer hooks when the person has nothing:

  • H-motion: a signature animation.
  • H-illus: original illustration, when open-library art would read generic.
  • H-motif: painted grounds and textures.

Everything is in this folder:

  • The bridge (scripts/paint.py, Python standard library, offline). It reads the design system and writes a brief, a palette and rules. It scaffolds film projects, checks scenes, and brings finished work into the app with accessibility safeguards and a ledger.
  • The painting engine (engine/, styles/, scripts/new-film.sh). It is vendored from generative-painting-skills (MIT, commit 1ff8212; THIRD-PARTY.md lists what changed). It plans shots, paints plates procedurally with p5.js and p5.brush, edits them with hard cuts, adds particle motion and a quiet score, and exports MP4, WAV and lossless PNG files that must pass automated checks.

Talk plainly: one idea and one question per message, and recommend one option with its reason. The first time a term appears, give its plain meaning plus "Designers: X · Code: Y" (references/rules.md in the opendesigner skill).

Run paint.py (scripts/paint.py) from the person's project root. Knowledge:

FileRead it when
references/flow.mdFirst: the whole process, what the person does and what exists after each step
references/brand-to-paint.mdExplaining or overriding any line of the brief
references/painting-workflow.mdPlanning and painting: shot list, plates, motion, sound, export (the engine's own guide)
references/choosing-and-mixing-styles.md, styles/<style>/STYLE.mdChoosing or mixing styles; how each material works
references/editing.md, drawing-kit.md, motion-in-the-medium.md, sound.md, widescreen-and-titles.md, verifying.md, experimenting.md, adding-a-style.mdThe step you are on
The images each styles/<style>/STYLE.md lists under "See it first" (styles/<style>/examples/, references/*.jpg)Before choosing a style: what each one looks like

Hard rules

  • Ask before installing. Each film project needs npm ci (p5, p5.brush, Vite and Playwright, every package pinned by engine/package-lock.json) and npx playwright install chromium. Both download code. paint.py scaffold prints them and never runs them: get a yes first. scripts/new-film.sh installs at once, so it is for people running it by hand.
  • Structure, never identity. The watercolour scene library reconstructs one earlier film; three ported sketches of unknown authorship are left out. Use the library for composition, rhythm and archetypes only. Every shipped plate needs this product's own subject and the brand palette. paint.py lint flags library scenes left in their original colours (guardrails.md).
  • The brand palette only. Scenes take colours from src/brand.js, one family per plate. Pigment mixing makes in-between colours; that is the medium, not drift.
  • No text painted into plates. Words go in a fixed-anchor phrase track, measured at 4.5:1 under the glyph mask (WCAG 2.2 AA is locked). The font must be licensed for rendering into video and images; never guess a licence.
  • Motion stays where motion belongs. Films and loops go in marketing, onboarding and celebration moments, never on functional screens (DC-L05-21, STD-when-to-animate-11; over 1 s only because it is illustrative, STD-easing-duration-09). In the app a film is:
    • muted;
    • shown as a still for prefers-reduced-motion;
    • given a pause control past 5 s (WCAG 2.2.2). paint.py ingest writes all three.
  • Determinism is the engine's contract. All randomness goes through the seeded kit, never Math.random. Plates are painted once and held. Integer frames are the source of truth.
  • Report what was checked. Say what the export measured, and what you actually looked at or listened to.

1. Look

  • paint.py doctor: the engine and styles here, Node, npm, ffmpeg and Python.
  • Read DESIGN.md and opendesigner/state.json:
    • the dials;
    • the answers to Q-img-04, Q-img-06 and Q-img-07;
    • the H-motion, H-illus and H-motif hooks;
    • the product, its audience and its "memorable thing".
  • Without a design system: use --accent '#hex', and offer the opendesigner interview first.

2. Brief

python3 paint.py brief --slot hero|launch|onboarding|celebration|og-image|empty-state|background [--message "..."] [--style ...]

It writes opendesigner/paintings/<slot>/brief.md and brief.json:

  • the recommended style and why (from the dials, or Q-img-04);
  • whether mixing styles is allowed;
  • 7-9 palette families (one more per extra accent ramp) built only from the token ramps, each marked for legible text;
  • the font decision;
  • the slot's size and length (the engine paints 600x600 squares; other sizes need the compositor in references/widescreen-and-titles.md §2);
  • the motion, sound and identity rules.

Show the style and a line per family, then ask: this style, or another?

3. Scaffold (ask first)

python3 paint.py scaffold <film-dir> --style <style> [--id <film-id>]      engine + demo + src/brand.js + BRIEF.md; installs nothing
cd <film-dir> && npm ci --no-audit --no-fund && npx playwright install chromium        only after the person's yes

The demo scenes are placeholders. npm run export once confirms the machine works.

4. Plan and paint

Follow references/painting-workflow.md:

  1. Shot list: show it and get it approved.
  2. One scene per plate, started from the closest scene-library recipe, with colours from src/brand.js:
    import { family } from '../brand.js';
    const F = family('brand-night');
  3. npm run plates -- --only <id>, then look at each plate at full size.
  4. Motion and sound. Particle colours live in src/particles.js: recolour the palettes of the motions you use from src/brand.js (lint checks them).

For charcoal and sketch, set PLATES[id].material.paper to the brief's paper colour.

5. Check

python3 paint.py lint <film-dir>     brand colours (scenes and particles), seeded randomness, no text in plates, no unadapted library scenes, length
npm run export                       the engine's checks; it refuses to finish if one fails

Then do a visual review (references/verifying.md) and judge as opendesigner-illustrate/references/ranking.md asks. python3 tools/cut_audit.py out/<film> needs numpy and scipy.

6. Bring it into the product

python3 paint.py ingest <film-dir> --to src/components/brand --as react [--poster <plate-id>] [--label "..."]
python3 paint.py ingest <film-dir> --to public/art --as img

ingest does four things:

  • Refuses a failed export.
  • Copies the MP4 and a poster: a plate from npm run plates, or else the film's anchor frame for that plate.
  • Writes the component: muted, a reduced-motion still, and a pause control past 5 s. Decorative art gets aria-hidden; meaningful art gets --label.
  • Logs a ledger row and CREDITS.md.

Then:

  • Look at it in the running app, on phone and desktop, light and dark, with reduced motion on and off. The opendesigner-review skill does this with evidence.
  • engine.py set hooks.H-motion.status '"tool"' --why "painted with opendesigner-paint" (or H-illus), then engine.py review.

Hosts

  • Claude Code, Codex, Cursor: everything above, once the person agrees to the installs.
  • claude.ai: the sandbox usually cannot install Chromium or ffmpeg. Write the brief and the scenes there; paint and export on the person's machine.
  • No code: the brief works as a written commission for a person or a tool.

If something was missing, wrong or confusing, record it with engine.py feedback "..." --kind gap|bug|confusing|idea (references/improve.md in the opendesigner skill). Nothing is posted without the person's OK.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.