Agent Skills for Claude Code

A community skill library that turns design systems, Figma edits, and frontend craft into installable Claude Code workflows.

Stars
★ 106
Last updated
3d ago
License
MIT
Primary language
HTML

At a glance

How it runs
CLIAgent plugin / skill
Works with
Portable with changesClaude Code · Claude.aiChatGPT (Partial support)
Cost
Free software; you pay for model usage
Setup effort
Low · running in minutes
You'll need
npx skills CLIClaude 帳號(Customize → Skills 上傳 ZIP)Shell / CLINetwork accessLocal filesystem
Typical use
A frontend engineer inheriting an aging app runs the accessibility skill for a WCAG 2.2 audit and concrete screen-reader and keyboard fixes
Not a fit if
  • Teams expecting a published ChatGPT plugin install path; the ZIP is not one
  • Users who refuse to run any terminal or npx command
  • Teams wanting one monolithic tool instead of hand-picking individual skills
Maintenance
Commits in the last 30 dayslast commit 2026-11-18+1.0 ★/day over 7 days

What does this agent do, and when should you use it?

Agent Skills is a community-maintained repository from thedesignproject containing roughly 20 skills for Claude Code and other local coding agents. Skills cover accessibility auditing, design-system extraction and AI-readable component metadata, Figma read/write flows, frontend interface generation, PRD authoring, and prompt engineering. Each skill lives in its own directory (e.g. skills/accessibility/, skills/figma-use/) and activates either as a slash command or automatically when a request matches its description. Installation goes through the npx skills CLI, which offers an interactive picker, per-skill selection with -s, and global installs with -g; Build My Designer Brain additionally ships as a downloadable ZIP uploaded through Claude's Customize → Skills flow. The repository itself is content, not a runtime — execution happens inside whichever host agent loads the skill. It is MIT licensed and includes CONTRIBUTING.md and CODE_OF_CONDUCT.md.

The repo distributes capability by skill directory. accessibility audits and improves web accessibility to WCAG 2.2, covering screen-reader support and keyboard navigation. agentic-design-systems and ai-component-metadata generate machine-readable metadata for components, variants, tokens, and anti-patterns so an LLM knows when and how to use each one. extract-design-system pulls color, typography, and spacing primitives from a public website and emits starter token files. figma-use is a mandatory prerequisite context before any use_figma MCP call, while figma-generate-design pushes an existing app page from code into Figma. frontend-design builds distinctive interfaces, web-design-guidelines reviews UI code against Web Interface Guidelines, and presentation emits single-file HTML/CSS/JS slide decks. Workflow skills include create-prd (builds a PRD from a transcript, optionally creating Linear projects and issues), pr-branch-naming, business-kickoff-workshop (45-minute Business Model Canvas agenda), and product-context-builder (ties together AGENTS.md, CLAUDE.md, PRODUCT.md, DESIGN.md, and project skills). caveman offers lite/full/ultra compression levels cutting token usage ~75%, and create-skill plus writing-skills handle authoring and verifying the skills themselves.

  1. A frontend engineer inheriting an aging app runs the accessibility skill for a WCAG 2.2 audit and concrete screen-reader and keyboard fixes
  2. A design-system maintainer uses ai-component-metadata to produce structured metadata so coding agents pick the right variant
  3. A designer pushes a multi-section page from code into Figma with figma-generate-design instead of rebuilding it by hand
  4. A product manager feeds a customer transcript to create-prd and gets a technically scoped PRD, optionally synced to Linear
  5. An individual designer runs Build My Designer Brain to capture personal critique methods and AI working preferences via a guided interview
  6. A team uses product-context-builder to stand up AGENTS.md, CLAUDE.md, PRODUCT.md, and DESIGN.md context files in one pass

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • npx skills add thedesignproject/agent-skills
  • npx skills add thedesignproject/agent-skills -s figma-use
  • npx skills add thedesignproject/agent-skills -s figma-use -s frontend-design
  • npx skills add --help

How do you install or deploy this agent?

The recommended path is the npx skills CLI:

# Install all skills — interactive list of everything in the repo
npx skills add thedesignproject/agent-skills

# Install one skill
npx skills add thedesignproject/agent-skills -s figma-use

# Install several at once
npx skills add thedesignproject/agent-skills -s figma-use -s frontend-design

Add -g to install globally rather than into the current project; full flag list:

npx skills add --help

For a no-terminal route, Build My Designer Brain is distributed as a ZIP uploaded in Claude:

https://github.com/thedesignproject/agent-skills/raw/refs/heads/main/downloads/build-my-designer-brain.zip

For local coding agents, unzip the download and install the complete folder following the "Install the shared skill" section in guides/build-my-designer-brain.md.

How do you use this agent?

Skills are either slash commands or auto-triggered. Slash command examples:

/create-prd path/to/transcript.md
/presentation [topic]
/pr-branch-naming [feature description]
/caveman

These are typed directly into Claude Code, without an npx prefix. Auto skills fire on natural language such as "audit accessibility", "review my UI", or "find a skill for X". Customize company-context.md before using create-prd. Load figma-use before any use_figma MCP write or scripted read in a Figma file. For the Claude upload route: open Customize → Skills → + → Create skill → Upload a skill, select the ZIP, enable the skill and enable file creation, then say "Use Build My Designer Brain to help me build my personal designer brain."

What are this agent's strengths and limitations?

Pros
  • Around 20 skills split into separate directories, so you can install only what you need with -s instead of taking the whole bundle
  • figma-use codifies Figma write/read failure modes as a mandatory prior context, reducing broken use_figma MCP calls
  • figma-generate-design pushes an existing code page into Figma, eliminating manual redraw work
  • agentic-design-systems and ai-component-metadata target the specific problem of making a component library legible to an LLM
  • caveman quantifies savings with lite/full/ultra levels cutting roughly 75% of tokens
  • A ZIP download plus in-app Claude upload path lets non-terminal users install Build My Designer Brain
Limitations
  • The catalogue is oriented around Claude Code and Claude first; the README states the ZIP is not a published ChatGPT plugin and ChatGPT setup varies by surface
  • Many skills lean on third-party services (Figma MCP, Linear) or fetch external websites, so you supply credentials and connectivity
  • No dependency graph or version pinning between skills; when installing individually you must judge which companions are needed
  • The create-skill description refers to Cursor Agent Skills, inconsistent with the Claude Code focus elsewhere
  • No shared evaluation or regression data demonstrating each skill's output stability on real projects

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Agent Skills for Claude Code This agent 42 · Major gaps CLIFree + model costs ★ 106 3d ago HTML Claude Code · Claude.ai
Dembrandt UX Skills 60 · Some gaps Agent plugin / skillFree ★ 197 1d ago JavaScript Claude Code
Designpowers 62 · Some gaps Agent plugin / skill ★ 262 5mo ago Shell Claude Code
Superdesign Skill 56 · Major gaps Agent plugin / skillFreemium ★ 1k 3mo ago JavaScript Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
42/ 100 5-point scale 2.1 / 5
Trust 10/29
Reliability 5/14
Adaptability 10/18
Convention 8/18
Effectiveness 6/13
Verifiability 3/8
Why each dimension lost points
Trust10 / 29 · 1.7/5

build-my-designer-brain explicitly avoids searching unrelated personal directories, forbids unrequested Git init/publishing, and requires authorization plus path verification before writes, showing limited least-privilege and confirmation awareness (least_privilege=1, user_confirmation=2). However most skills (create-prd creating Linear projects, Figma writes, extract-design-system scraping sites) do not document external side effects or data flows in visible files, so data_flow_transparency and external_effects score 1. Sensitive-data handling appears only as a 'do not copy restricted company material' line in one skill, scoring 1. Dependency security has no evidence (npx install, python scripts, MCP calls with no pinning or audit), scoring 0. Rollback is only implied by draft-save/verify-write behavior with no undo mechanism, scoring 1. Source attribution appears only for the accessibility skill (author + MIT); others lack provenance, scoring 1.

Reliability5 / 14 · 1.8/5

Individual SKILL.md files are internally coherent, but README labels create-skill as Cursor-oriented and overlaps with writing-skills, a minor inconsistency, self_consistency=1. Dependency availability: npx skills, python scripts, and Figma MCP are external dependencies with no availability or version constraints, scoring 1. Failure messages and error handling are largely absent from visible files, scoring 1.

Adaptability10 / 18 · 2.8/5

README and skills describe scenarios, triggers, and when not to use (e.g. agentic-design-systems excludes purely human-facing work) reasonably well, audience_and_scenarios and capability_boundaries score 2. Trigger precision is clear with Auto/Slash distinction and trigger phrases, scoring 2. Environment fit mentions Claude, ChatGPT, and local coding agents, but ChatGPT is explicitly 'not a published plugin', so cross-environment support is incomplete, scoring 1.

Convention8 / 18 · 2.2/5

README provides a skill table, guides table, and install instructions, so information_architecture and install_notes score 2. Naming is consistent between skill names and directories, but version fields vary in format (1.1, 1.0.0, missing), scoring 1. Examples are plentiful but there is no FAQ, scoring 1. Known limitations are only mentioned sporadically, scoring 1. LICENSE is full MIT, scoring 2. No CHANGELOG or release history, scoring 0. Maintenance responsibility points only to CONTRIBUTING/Code of Conduct and publisher identity is unverified, scoring 1.

Effectiveness6 / 13 · 2.3/5

Skill outputs are mostly structured templates and workflows that are directly usable, output_usability=2. But many skills are generic design/prompt knowledge overlapping public resources, limiting marginal value to 1. Cost-benefit: some skills (e.g. caveman token savings) have clear benefit, but overall lacks quantification or comparison evidence, scoring 1.

Verifiability3 / 8 · 1.9/5

Most README capability claims map to specific SKILL.md files, claim_traceability=1; but assertions like '~75% token reduction' and 'prevents common failures' are unsupported. Cross-source corroboration is limited to README-skill cross-references, scoring 1. Fact and inference are mostly separated, but some skills state recommendations as definitive conclusions, scoring 1.

Risks and how to mitigate them
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Publisher identity is unverified by FollowAgents; maintenance responsibility and update path are unclear.
  • External dependencies (npx install, python scripts, Figma MCP) have no version pinning or security audit notes.
  • Skills such as create-prd (Linear project creation), Figma writes, and website scraping do not document external side effects or data flows in visible files.
  • No CHANGELOG or release history, and skill version fields use inconsistent formats.
  • Some capability claims (e.g. ~75% token reduction) lack traceable evidence.
Evidence confidence: Low Reviewed Nov 15, 2026 Reviewed revision f579d56b99a5 New commits since this review; the score may not cover them
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/thedesignproject-agent-skills.svg)](https://followagents.com/en/agents/thedesignproject-agent-skills)
<a href="https://followagents.com/en/agents/thedesignproject-agent-skills"><img src="https://followagents.com/badge/thedesignproject-agent-skills.svg" alt="FARS review"></a>

FAQ

Do skills activate on their own or only when I call them?
It depends. The README's "How to use" column marks skills such as accessibility and frontend-design as Auto, while create-prd, presentation, pr-branch-naming, and caveman require an explicit slash command.
What does the Figma work require beyond installation?
You need Figma MCP wired into your host with access to the target file; figma-use is explicitly a mandatory prerequisite before any use_figma MCP call, added to avoid common read/write failures.
Can I install the ZIP in ChatGPT?
No. The README states the ZIP is not a published ChatGPT plugin and that ChatGPT installation differs by surface; the ZIP upload flow is for Claude's Customize → Skills path.
Do these skills touch my filesystem or the network?
The skills are prompt and workflow files; the host agent performs the actions. extract-design-system fetches a public website and create-prd reads your codebase and can create Linear issues, so network and file access depend on which skills you enable.
Can I adopt just one or two skills?
Yes. npx skills add thedesignproject/agent-skills -s <skill-name> installs a single skill, multiple -s flags install several at once, and -g makes it global.
Is Agent Skills for Claude Code free?
The software is free, but Agent Skills for Claude Code calls an AI model, and you pay that model provider for usage (or run a local model).
Is Agent Skills for Claude Code open source?
Yes. The source code is on GitHub (thedesignproject/agent-skills) under the MIT licence.
Is Agent Skills for Claude Code safe to use?
FollowAgents' FARS review (Nov 15, 2026) scored Agent Skills for Claude Code 10/29 on trust and safety. The review did not find these safety controls in the source: dependency security. Before giving it real accounts or data: Pin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container. The repository has changed since that review.
Is Agent Skills for Claude Code still maintained?
Yes. The latest commit on GitHub is from 2026-11-18, within the last 30 days.
Who makes Agent Skills for Claude Code?
Agent Skills for Claude Code is published on GitHub by the account thedesignproject. FollowAgents has not verified which person or company is behind that account.
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts