StyleSeed

A skill set that locks a project's design decisions into rules and checks the UI with a score gate and rendered screenshots

SkillMedium risk

bitjaru/styleseed

Install

npx skills add bitjaru/styleseed

Installs the 23 core skills. Then run /ss-setup in Claude Code or $ss-setup in Codex.

This is third-party code. Review the repository files before installing.

What it does

StyleSeed turns agreed design decisions into project-local rules: accent color, radius, font and motion are written to STYLESEED.md and re-read for every visual task. The set includes the styleseed router and 22 ss-* skills for setup, screen building, audits, code scoring, render verification, references and palette generation. The ss-resolve skill compiles a small task-specific rule bundle from the full handbook, and ss-score rates a screen from 0 to 100 and sends it back for revision below the threshold. Node.js scripts for palette generation and checks are included.

Who it is for. For frontend developers and designers who build UI with an agent and want consistency across screens.

Good fit when

  • Agent-built UI looks generic and each screen drifts from the style
  • You need to lock chosen colors, fonts and radii for the whole project
  • You want a finished screen checked against rules and a screenshot before review

Not a fit when

  • The team already has a strict design system with its own checks
  • The project is not web-based and there is nowhere to run the Node.js scripts

Example request

Set up StyleSeed for this project and build the billing settings page with the score gate

Limitations

Default rules target React, Tailwind and shadcn/ui, so some checks do not apply to other stacks. The authors state that the score does not replace a designer's acceptance. The full rule set is large, and without ss-resolve the agent spends a lot of context.

How to disable. Remove the styleseed and ss-* folders from the agent's skills directory and delete STYLESEED.md and the .styleseed folder from the project.

Security check

  • The skills run Node.js scripts and edit UI files in the project

README in short

The README suggests installing via npx skills and pasting a setup message to the agent. It describes rules that catch typical signs of generated UI, the render, score and revise loop, and a benchmark on rendered screens. It lists output grammars, brand recipes, palettes and skins. There are separate guides for Claude Code and Codex. MIT licensed.

SKILL.md

---
name: styleseed
description: Route a StyleSeed request to exactly one first workflow after resolving the current artifact boundary. Use when the user asks generally for StyleSeed help rather than invoking one specific ss-* skill.
argument-hint: "[request or goal]"
allowed-tools: Read, Grep, Glob, Bash
---

# StyleSeed router
## Registry-first artifact boundary

When `.styleseed/project.json` and `.styleseed/artifacts/index.json` exist, resolve the requested artifact ID first, then read only `.styleseed/bundles/<artifact-id>.md` and `.styleseed/manifests/<artifact-id>.json`. Never fall back to the global legacy bundle for a registry project. Legacy projects may use `.styleseed/effective-rules.md` only when no registry exists.

This router chooses exactly one first workflow. It is the primary entry skill when a user asks for
StyleSeed help in general terms.

## First principles

- Help coding agents repeat expert design decisions, not replace expert authority with presets.
  Preserve approved project choices; identify unsupported or unresolved choices before routing
  to work that would change them. A score is not human acceptance.
- Resolve the current artifact first when `.styleseed/project.json` and `.styleseed/artifacts/index.json`
  exist.
- Never fan out to “run every StyleSeed skill.”
- Never copy the full design handbook into this router. Route to the maintained skill that already
  owns the workflow.
- Learning is an optional extension, not part of the core install. Only after an explicit capture
  request such as “remember this lesson” may the router check whether `ss-learn` is separately
  installed. If absent, state the dependency precisely; never auto-install it.
- If the request is ambiguous, ask one bounded clarification question that decides the first workflow.

## One-workflow routing table

Choose exactly one first workflow:

- `ss-setup` for first-time method setup, missing lock/setup, or “set up StyleSeed for this project.”
- `ss-build` for implementing or redesigning a concrete screen/component/page when the direction is already known.
- `ss-reference` for compiling supplied references that are not already represented by a maintained grammar.
- `ss-studio` for creative direction, exploration, concept generation, or multi-direction concept work.
- `ss-audit` for critique, heuristic review, UX issue finding, or “what is wrong with this screen.”

FAQ

How does ss-studio differ from ss-build?

ss-studio proposes several creative directions and waits for a human choice, while ss-build implements a screen for an already agreed direction.

Can I use it without installing?

You can give the agent the project's llms.txt, but rule compilation and checks will not run.

Editors’ pick

Official shadcn/ui skill and MCP server: the agent searches registries, adds components via the CLI and follows the library's rules

SkillMedium risk123.8K
Editors’ pick

A frontend design skill with 23 commands, live browser iteration and 61 detector rules for common AI-generated UI tells

SkillHigh risk68K
Editors’ pick

Vercel's skills: React and Next.js performance, UI guideline audits, React Native, component composition and deploys

SkillMedium risk31.2K
Editors’ pick

An MCP server that gives the agent simplified Figma layout data from a link so it can build the screen more accurately than from a screenshot

MCP serverMedium risk15.9K