SkillVault

License Skills PRs Welcome Validate Skills Stars Last Commit

A curated, tested collection of Claude Skills — vetted, trimmed, and confirmed to work in claude.ai chat (not just Claude Code).

See CONTRIBUTING.md for the vetting checklist, SECURITY.md for how we handle unsafe skills, FAQ.md for common questions, and CHANGELOG.md for version history.

A machine-readable index of every skill — name, description, category, compatibility, license, source — is available in catalog.json.

Using Claude Code instead of claude.ai? Run ./scripts/install-claude-code.sh --list to see everything, or ./scripts/install-claude-code.sh <name> to install specific skills straight into ~/.claude/skills/.

Contents

Why this exists

Most "awesome-claude-skills" lists are raw link dumps. A Snyk security audit of ~4,000 skills found 36% had at least one security flaw and 76 were confirmed malicious. Separately, a lot of skills are written for Claude Code specifically — they lean on subagent dispatch, persistent local git state, or CLI tools like gh — and silently do nothing in claude.ai's web/mobile chat interface, where the sandbox resets between sessions.

This repo exists to fix both problems:

  • Every skill here has been read in full, not just linked.
  • Every skill has been trimmed of Claude-Code-only dependencies and bloat (one skill went from 1.4MB → 95KB by dropping 17 irrelevant language guides).
  • Every skill's frontmatter has been checked against claude.ai's upload requirements (only name, description, license, compatibility, metadata, allowed-tools are allowed — extra fields like version/author/tags will reject the upload).
  • Every skill is labeled by where it actually works.

How to use a skill

claude.ai (web/mobile): Settings → Capabilities (or Skills) → Upload skill → pick the .zip from downloads/ → toggle on. Requires a paid plan with Code execution enabled.

Claude Code: cp -r skills/<name> ~/.claude/skills/ or clone this repo and point your skills directory at it.

Compatibility legend

Badge Meaning
🟢 Chat-ready Pure markdown guidance, no CLI/browser/local-git dependency. Works fully in claude.ai chat.
🟡 Chat-partial Core guidance works in chat; some referenced scripts/tools need a local environment to run.
🔵 Claude Code Needs persistent local state, gh CLI, or subagent dispatch. Chat-incompatible by design.

Engineering discipline

Skill Compat What it does Source License
test-driven-development 🟢 Red-green-refactor discipline; forces a failing test before implementation code obra/superpowers MIT
systematic-debugging 🟢 Root-cause tracing over guess-and-check; condition-based waiting instead of sleeps obra/superpowers MIT
verification-before-completion 🟢 Requires running and confirming output before claiming something is fixed/done obra/superpowers MIT
code-review-skill 🟢 Review guidance for React/TS/Python/FastAPI + architecture, security, performance checklists (trimmed from 21 language guides to 4) awesome-skills/code-review-skill MIT

Interview / system design

Skill Compat What it does Source License
system-design 🟢 HLD/LLD interview coach — estimation, scaling, caching, CAP/PACELC trade-offs git-akki/system-design-skill MIT (Akash Yadav)
tech-resume-optimizer 🟢 Resume optimization specifically for SWE/PM/technical roles Paramchoudhary/ResumeSkills MIT
resume-bullet-writer 🟢 Turns weak bullets into metric-driven, impact-focused statements Paramchoudhary/ResumeSkills MIT
interview-prep-generator 🟢 Generates STAR stories and practice questions from a resume Paramchoudhary/ResumeSkills MIT

Design / frontend

Skill Compat What it does Source License
design-studio 🟢 Forces real brand/context gathering before designing; pushes against generic "AI slop" UI (gradient blobs, default fonts); ships multiple design variations. Renamed from claude-design — claude.ai blocks skill names containing the reserved word "claude". jiji262/claude-design-skill MIT (jiji262)
web-design-guidelines 🟢 Accessibility/UX compliance checklist from Vercel Engineering vercel-labs/agent-skills MIT
vercel-react-best-practices 🟢 45 React/Next.js performance rules (data fetching, caching, bundle size), organized by impact. Trimmed: dropped a 108KB Claude-Code-only AGENTS.md bootstrap file. vercel-labs/agent-skills MIT

AI/ML

Skill Compat What it does Source License
pinecone-rag 🟢 Production RAG patterns for Pinecone — hybrid search, metadata filtering, namespaces Orchestra-Research/AI-Research-SKILLs MIT
langchain-agents 🟢 LangChain patterns — ReAct agents, tool calling, memory management, vector retrieval Orchestra-Research/AI-Research-SKILLs MIT

DevOps / deployment

Skill Compat What it does Source License
ci-cd-pipelines 🟡 GitHub Actions/GitLab CI/Jenkins pipeline design, security, secrets management. Note: upstream SKILL.md references a scripts/generate_pipeline.py helper that doesn't actually exist in the source repo — guidance and templates work, that one script doesn't. compufreq/claude-skills CC BY-NC-SA 4.0 (Alaa Alhorani) — ⚠️ non-commercial, share-alike, attribution required

Planning & workflow (from obra/superpowers)

Skill Compat What it does Source License
brainstorming 🟢 Structured idea generation before committing to an approach obra/superpowers MIT
receiving-code-review 🟢 How to actually process and act on review feedback instead of just reacting to it obra/superpowers MIT
writing-plans 🟡 Structured technical planning before implementation. Some sections reference subagent dispatch/worktrees — core planning guidance still applies solo. obra/superpowers MIT
executing-plans 🟡 Disciplined execution of a written plan, checkpoint by checkpoint. References subagents/worktrees in places — usable single-threaded. obra/superpowers MIT
requesting-code-review 🟡 How to frame a review request so you get useful feedback. Some subagent-dispatch content doesn't apply in chat. obra/superpowers MIT
writing-skills 🟡 Meta-skill: how to write a good SKILL.md yourself. Long (679 lines), some subagent-specific sections. obra/superpowers MIT

AI/ML — agent frameworks & tooling

Skill Compat What it does Source License
llamaindex 🟢 Data framework for RAG — 300+ ingestion connectors, indexing, query engines Orchestra-Research/AI-Research-SKILLs MIT
crewai 🟢 Multi-agent orchestration — role-based agent teams, sequential/hierarchical workflows Orchestra-Research/AI-Research-SKILLs MIT
whisper 🟢 OpenAI's speech-to-text model — 99 languages, transcription, translation Orchestra-Research/AI-Research-SKILLs MIT
clip 🟢 Vision-language model — zero-shot image classification, image-text matching Orchestra-Research/AI-Research-SKILLs MIT
instructor 🟢 Structured, type-safe data extraction from LLM outputs with Pydantic validation Orchestra-Research/AI-Research-SKILLs MIT
dspy 🟢 Declarative LM programming — automatic prompt optimization, modular RAG/agent pipelines Orchestra-Research/AI-Research-SKILLs MIT
langsmith 🟢 LLM observability — tracing, evaluation, production monitoring Orchestra-Research/AI-Research-SKILLs MIT

Career / job search (additional)

Skill Compat What it does Source License
linkedin-profile-optimizer 🟢 Optimize LinkedIn for recruiter search visibility and engagement Paramchoudhary/ResumeSkills MIT
cover-letter-generator 🟢 Personalized cover letters from resume + job description Paramchoudhary/ResumeSkills MIT
resume-tailor 🟢 Customize resume per job posting while staying truthful Paramchoudhary/ResumeSkills MIT
resume-ats-optimizer 🟢 ATS compatibility check + keyword match analysis Paramchoudhary/ResumeSkills MIT
resume-formatter 🟢 ATS-friendly formatting, clean scannable layouts Paramchoudhary/ResumeSkills MIT

Testing

Skill Compat What it does Source License
vitest 🟢 Vitest framework — fast unit tests, Vite integration, component testing, mocking full-stack-skills/testing-skills Apache-2.0
pytest 🟢 pytest framework — fixtures, parametrization, mocking, plugins full-stack-skills/testing-skills Apache-2.0
playwright 🟢 Cross-browser E2E testing — browser automation, page objects full-stack-skills/testing-skills Apache-2.0

Database / backend

Skill Compat What it does Source License
docker 🟢 Container build/security/deployment best practices Mindrally/skills Apache-2.0
mongodb-development 🟢 MongoDB with Mongoose, aggregation pipelines, TypeScript patterns Mindrally/skills Apache-2.0
supabase-development 🟢 Supabase database, auth, real-time subscriptions, Edge Functions Mindrally/skills Apache-2.0
prisma-client-api 🟢 Prisma Client query API — CRUD, transactions, raw SQL (targets Prisma 7.9.x) prisma/skills (official) MIT
prisma-database-setup 🟢 Configuring Prisma across PostgreSQL/MySQL/SQLite/MongoDB providers prisma/skills (official) MIT

Security

Skill Compat What it does Source License
auth-implementation-patterns 🟢 JWT, OAuth2, session management, RBAC — secure auth system design wshobson/agents MIT
secrets-management 🟢 Secure secrets handling in CI/CD — Vault, AWS Secrets Manager, platform-native wshobson/agents MIT

What we deliberately left out

  • dispatching-parallel-agents, subagent-driven-development, using-git-worktrees (from obra/superpowers) — depend on Claude Code's subagent dispatch and persistent local git worktrees. Dead weight in claude.ai chat.
  • Most GitHub PR-review skills (aidankinzett/claude-git-pr-skill, SpillwaveSolutions/pr-reviewer-skill, maddevsio/ai-review-skill) — all require an authenticated local gh CLI.
  • career-ops and its forks — need Playwright browser automation and persistent local application-tracking state.
  • Most "MERN stack" skill searches — turned out to be course/roadmap content, not actual Skills.
  • The rest of obra/superpowers (using-superpowers, finishing-a-development-branch, dispatching-parallel-agents, subagent-driven-development, using-git-worktrees) — either meta-skills whose only job is invoking other skills, or too deeply built around Claude Code's subagent/worktree model to be honestly labeled 🟢 or even 🟡.
  • ~90 skills in Orchestra-Research/AI-Research-SKILLs covering distributed training, fine-tuning (LoRA/QLoRA), quantization, and mechanistic interpretability — genuinely excellent if you're training or fine-tuning your own models, but not relevant to building applications on top of hosted LLM APIs, which is what this repo's audience is actually doing.
  • ~16 unused skills in Paramchoudhary/ResumeSkills (academic-cv-builder, executive-resume-writer, offer-comparison-analyzer, salary-negotiation-prep, reference-list-builder, career-changer-translator, etc.) — aimed at later-career or post-offer stages, not the active-application stage this repo targets.
  • 6 of 9 skills in full-stack-skills/testing-skills (jest, junit, cypress, selenium, appium, detox) — overlap with what's already covered (Vitest > Jest for modern Vite projects), or target Java/mobile-native stacks outside this repo's web/AI focus.
  • Jest and Cypress specifically — kept only Vitest and Playwright since they're the more current choice for Vite-based projects and cross-browser E2E respectively; including near-duplicate frameworks doesn't add real value.
  • ~230 of 240+ skills in Mindrally/skills — a converted-from-Cursor-rules mega-repo covering Kubernetes, Terraform, every cloud provider, every ORM, every testing framework. Only pulled Docker/MongoDB/Supabase since those match this repo's actual stack; the rest is enterprise-infra territory (Kubernetes, multi-cloud IaC) outside scope.

Contributing

Found a skill that's genuinely useful and works in claude.ai chat (not just Claude Code)? Open a PR with:

  1. The skill folder under skills/<name>/
  2. A zipped copy under downloads/<name>.zip
  3. A row in the relevant README table with compat badge + source attribution

Please don't submit skills you haven't personally read in full — see the Snyk audit linked above for why that matters.

License

This repo's own content (README, categorization, trimming) is MIT. Each skill retains its original upstream license — see the table above and each skill folder for details. This is a redistribution/curation project, not original authorship of the skill content itself.