design-builder
AGENTS.md4 starsChanged 13 days agoMITView on GitHub
This repository is a universal Claude Code and Codex plugin: design-builder. It bundles 6 slash commands, the design knowledge skill plus sister knowledge skills (motion, playwright-cli, figma-use), and the design-auditor sub-agent. Manifests: .claude-plugin/plugin.json and .codex-plugin/plugin.json. Portable entry: skills/design-builder/SKILL.md. Shared host rules: references/host-compatibility.md. Knowledge skill: skills/design/SKILL.md. Commands: commands/. Sub-agent: agents/design-auditor.md. When the plugin is installed, users invoke commands as /design-builder:<command> — setup, start (alias of setup), design_page, design_screen, build, improve, review. Every command must: (1) resolve facts through Layer 1 in…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,700 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- design-builderSkill
- designSkill
- figma-useSkill
- motionSkill
- playwright-cliSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 7
When the plugin is installed, users invoke commands as /design-builder:<command> — setup, start (alias of setup), design_page, …
- Very specific: line 45
Every command ends with a Next: block recommending the next concrete action.
- Very specific: line 61
Keep native Claude commands and the auditor adapter.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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.
Files
design-builder Plugin
This repository is a universal Claude Code and Codex plugin: design-builder. It bundles 6 slash commands, the design knowledge skill plus sister knowledge skills (motion, playwright-cli, figma-use), and the design-auditor sub-agent.
Manifests: .claude-plugin/plugin.json and .codex-plugin/plugin.json. Portable entry: skills/design-builder/SKILL.md. Shared host rules: references/host-compatibility.md. Knowledge skill: skills/design/SKILL.md. Commands: commands/. Sub-agent: agents/design-auditor.md.
When the plugin is installed, users invoke commands as /design-builder:<command> — setup, start (alias of setup), design_page, design_screen, build, improve, review.
Structure
.claude-plugin/plugin.json— Claude Code manifest (name, version, mcpServers via.mcp.json, commands at./commands/).claude-plugin/marketplace.json— single-plugin marketplace (so the repo itself is installable as a marketplace source).mcp.json— shared MCP server config (designlib+figma)commands/— 6 user-facing commands:setup.md,start.md(alias),design_page.md,design_screen.md,build.md,improve.md,review.mdagents/design-auditor.md— sole sub-agent. Used by/review. Returns P0-P3 report + JSON contract; supportsvisual/code_only/mixedmodes; runs the AI-slop detector. Auto-captures running web apps viaplaywright-cliwhen the caller passesurls.skills/design/SKILL.md— knowledge skill: 3-layer model, Layer 2 filters (incl. Distinctiveness Gate), Layer 1 source order,get_design_reference()resolver (defined inreferences/layer1-resolvers.md), reference index.skills/design/references/— deep documentation (architecture, layer1-resolvers, inspiration_pages, commands, distinctiveness-gate, design-dials, designlib-mcp, ux-writing, web/, ios/, figma/, brand/, design/, slides/, system/, ui-styling/)skills/design/data/— CSV databases for BM25 searchskills/design/scripts/— Python (search, design_system, system preview) + Node (anti-pattern detector). Stdlib only.skills/design/templates/— starters (iOS SwiftUI theme, brand-guidelines starter)skills/motion/— sister knowledge skill. Web-UI animation library knowledge (GSAP, WAAPI, CSS, Anime.js, Lottie, Three.js) + HyperFrames short-video pipeline. Loaded bydesignand the relevant commands.skills/playwright-cli/— sister knowledge skill. Browser automation via@playwright/cli. Used by/review(auto-screenshots of running web apps) and/improve --restructure(captures live inspiration_pages references). Runtime:npx @playwright/cli, no local install.skills/figma-use/— Figma Plugin API wrapper (mandatory prerequisite for anyuse_figmawrite).NOTICE.md— attribution for 6 source open-source projects + Figma MCP guideLICENSE— MITCHANGELOG.md— version history
Architecture (three-layer rule)
Every command must: (1) resolve facts through Layer 1 in the fixed order (project tokens → designlib MCP → local CSV → iOS HIG → free generation), via get_design_reference(type, filters); (2) pass candidate output through all six Layer 2 filters (Direction, Dials, Aesthetics, Anti-Patterns, Distinctiveness Gate, Output Rules); (3) emit. Silently skipping Layer 2 is the #1 cause of generic "AI slop" output.
Extension markers in code: KB-EXTENSION (Layer 1 source), FILTER-EXTENSION (new Layer 2 filter). Pipeline markers from v1.2 are removed — there are no pipelines.
The 6 commands at a glance
| Command | Output target |
/design-builder:setup | <project>/design/design-system.md, style-guide.md, content-library.md, tokens.css, preview.html, references/, .cache/interview.json |
/design-builder:design_page <name> | <project>/design/pages/<name>.md (web page spec, no code) |
/design-builder:design_screen <name> | <project>/design/screens/<name>.md (iOS / native screen spec, no code) |
/design-builder:build [target] | Source code in user's source tree from a spec in design/pages/ or design/screens/ |
/design-builder:improve [target] | Patches user files via Edit; --restructure mode rebuilds sections from fresh anchors and may capture live inspiration_pages |
/design-builder:review [target] | <project>/design/reviews/review-YYYY-MM-DD-HHMM.md (delegates to design-auditor); accepts file path / Figma URL / HTTP(S) URL — URLs auto-captured via playwright-cli |
Every command ends with a Next: block recommending the next concrete action.
inspiration_pages (new in v2.0)
Primary Layer 1 source for whole-page references. Query current records via mcp__designlib__list_inspiration_pages / get_inspiration_page / list_inspiration_page_facets. Schema map: skills/design/references/inspiration_pages.md. Resolver contract: skills/design/references/layer1-resolvers.md.
Critical contract notes:
- MCP filters are SINGULAR strings (one
mood, onesignature, onekeyword). To combine values, call multiple times and dedupe byid. - List response returns a SUBSET of fields; deep-fetch via
get_inspiration_page(page_id=<id>)for full payload (palette, typography, sections, generation_prompt, why_it_works). - The deep-fetch param is
page_id, notid. - inspiration_pages are web-only; iOS / cross requests fall back to
landing_patternsor HIG.
Future: inspiration_parts (hero/CTA/paywall/pricing_table) — interface reserved in MCP_TOOL_MAP. Until then, partial-page references can be filtered today via good_for_stage='hero_section' | 'cta_band' | ... on inspiration_pages.
Compatibility and validation
Keep native Claude commands and the auditor adapter. Codex enters via
skills/design-builder/SKILL.md and reads the same command recipes. Do not fork
workflow content by host. Preserve the shared MCP configuration and resolve plugin
resources independently of the user's working directory. Delegation is optional.
Python utilities and the Node detector's --fast mode require no third-party
packages. Browser capture, DOM/URL analysis and video workflows have optional
runtime dependencies; check live CLI help and installed versions.
Run python3 scripts/validate_plugin.py and
python3 -m unittest discover -s skills/design/scripts -p 'test_*.py'.
Validate Claude with claude plugin validate . and
claude plugin validate .claude-plugin/plugin.json.
For Codex, install the local marketplace and exercise $design-builder in a
separate fixture project, confirming it loads cached plugin files and writes only
project artifacts. Installation instructions are in README.md. Keep manifest
versions and changelog aligned; preserve LICENSE and NOTICE.md attribution.
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
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.

