design-skills

Design skills for Claude Code, with the whole toolchain in a versioned, headless container: Blender, GIMP, Inkscape, Krita, Scribus, darktable, MLT, ffmpeg, ImageMagick, Ghostscript, FontForge, the web-image optimizers and a curated font library. Install the plugin, have Docker or Podman, and every skill works the same on Linux, macOS (Intel and Apple Silicon) and Windows (WSL2) — no per-machine setup.

Skills are installed under the codeonym-oss: prefix.

Skill	Covers	Image
codeonym-oss:design-sandbox	The environment itself: ds CLI, images, mounts, fonts, GPU, troubleshooting	—
codeonym-oss:design-doctor	Health check of everything; which skill/tool for which job	—
codeonym-oss:graphic-design	Logos, banners, social sizes, SVG export, vectorizing, web-image optimization	core
codeonym-oss:illustration	Krita canvases from presets, headless .kra export, tablet setup	full
codeonym-oss:photo-editing	darktable batch development, web-ready photos, metadata/GPS stripping, contact sheets	core · full
codeonym-oss:ui-design	Favicons/PWA icon sets + manifest, OKLCH color scales → CSS/Tailwind v4, design tokens	core
codeonym-oss:typography	Font inventory, specimens (any script), project fonts, WOFF2 + @font-face	core
codeonym-oss:print-layout	Scribus templates with bleed/CMYK, PDF preflight, proofs, compression, CMYK conversion	core · full
codeonym-oss:3d-modeling	Blender renders, design mockups, turntables, GLB (+Draco) conversion, STL	full
codeonym-oss:video-editing	Platform encodes, loudness normalization, GIF/WebM, thumbnails, headless Kdenlive/MLT renders	core · full
codeonym-oss:screen-recording	OBS setup on any OS, recording workflow, post-production of recordings	host · core

Install

claude plugin marketplace add codeonym-oss/claude-plugins
claude plugin install codeonym-oss@codeonym-oss

The codeonym-oss marketplace now lives in codeonym-oss/claude-plugins, beside the org's other plugins. If you added it from this repo before, switch once: claude plugin marketplace remove codeonym-oss, then the two commands above.

Then pull an image once (or let the first script pull it). ds ships inside the plugin; put it on your PATH:

ln -s ~/.claude/plugins/cache/codeonym-oss/codeonym-oss/*/bin/ds ~/.local/bin/ds
ds pull core     # ≈1.2 GB: graphic, photo, UI, type, print PDFs, video
ds pull full     # ≈2.9 GB: adds Blender, GIMP, Krita, Scribus, darktable, melt

Or just ask Claude: "check my design setup" runs the doctor.

How it works

bash skills/graphic-design/scripts/optimize.sh ~/Pictures/hero.png
        │  lib/env.sh → lib/ds.sh
        ▼
docker run --rm --user $UID:$GID \
  -v $PWD:$PWD -w $PWD                     ← your folder, same path inside
  -v <plugin>:<plugin>:ro                  ← the scripts
  -v ~/Pictures:~/Pictures                 ← any other folder you named
  [--device /dev/dri]                      ← VAAPI when the host has a GPU
  ghcr.io/codeonym-oss/design-skills:0.1.0-core  bash …/optimize.sh ~/Pictures/hero.png

• Each script re-runs itself in the image pinned to the plugin's VERSION, with identical paths, so outputs land next to your files and are owned by you.
• Scripts marked # ds-image: full use the full image; once the full image is pulled it serves everything.
• DS_RUNTIME=native runs on your own installed tools instead; ds doctor --verbose prints install commands for apt, pacman, dnf or Homebrew.
• Fonts in ./fonts are available to every tool in the container.

Desktop apps vs. the container

Everything scriptable runs headlessly in the image. Interactive desktop apps stay on your machine, and the skills explain them per OS: OBS and its plugins, keystroke overlays and auto-zoom recorders (screen-recording), Penpot, Figma and Lunacy (ui-design), color pickers, AI upscalers (Upscayl), GIMP plugins like Resynthesizer, the Kdenlive editor (projects are rendered headlessly with melt), and tablet drivers.

Images

Tag
ghcr.io/codeonym-oss/design-skills:0.1.0-core / :0.1.0-full	immutable release (what the plugin uses)
:0.1-core, :core, :0.1-full, :full, :latest	moving tags
:edge-core, :edge-full	latest main

Every published image carries an SBOM and SLSA provenance (docker buildx imagetools inspect <ref> --format '{{json .SBOM}}').

Built on Ubuntu 26.04 LTS (base pinned by digest), for linux/amd64 and linux/arm64. Publishing from a fork? GHCR creates new packages as private — set the package to Public once (package settings) so docker pull works anonymously.

Without Claude

docker run --rm -v "$PWD:/work" -w /work --user "$(id -u):$(id -g)" \
  ghcr.io/codeonym-oss/design-skills:0.1.0-core ds graphic-design/social-sizes banner.png
docker run --rm -it -v "$PWD:/work" -w /work ghcr.io/codeonym-oss/design-skills:0.1.0-full   # a shell with everything

Development

make build        # build core + full, tagged exactly as ds expects
make test         # unit tests on the host + integration tests inside both images
make lint         # shellcheck
make test-bash32  # host-side code under bash 3.2 (macOS)

Layout:

.claude-plugin/   plugin.json · marketplace.json
bin/ds            the CLI
lib/              ds.sh (dispatcher) · env.sh (sourced by every script) · check.sh + packages.tsv (checks, install hints)
skills/<name>/    SKILL.md · scripts/ · references/
docker/           Dockerfile (core-tools → full-tools → core / full)
tests/unit        bats: dispatcher, CLI, check library, repo consistency (fake container runtime)
tests/integration bats: every script against generated fixtures, inside the image (tag `full` = full image only)

See CONTRIBUTING.md for adding tools and skills and the release flow, CHANGELOG.md, and SECURITY.md to report vulnerabilities. The roadmap lives in the issues and milestones; bug reports and feature requests are welcome as GitHub issues.

Origins

Started as a copy of the author's personal Arch Linux workstation skills (codeonym-org/codeonym-arch-design), rebuilt here to be distro-agnostic and container-first.

License

MIT. The images bundle third-party open-source software under their own licenses (GPL, LGPL, OFL, Apache, …).