convez.ai
SkillIntegração & APIsIntermediário
Leon-Drq

OpenAgentSkill

Leon-Drq/openagentskill

Registry de skills para agentes IA com auditoria e recomendações

Ver no GitHub
BomPluginScore · de 0 a 100

Registro público e infraestrutura de descoberta, auditoria e recomendação de skills para agentes de IA. Funciona como um npm para skills de agentes, oferecendo busca, comparação, auditoria de segurança e APIs amigáveis para agentes como Claude Code, Cursor e Codex encontrarem e instalarem skills confiáveis automaticamente.

Instalação

$npx convez-ai Leon-Drq/openagentskill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
372
Forks
30
Acompanhando
8
Saúde
71%
Último commit
há 2 dias
Versão
cli-v0.3.0
Issues abertas
24
Contribuidores
4

Casos de uso

  • Descobrir e comparar skills reutilizáveis para tarefas específicas
  • Auditar segurança e qualidade de skills antes da instalação
  • Publicar e tornar skills descobríveis por agentes e desenvolvedores
  • Consultar programaticamente skills recomendadas via API

Veredito

Solução madura e bem estruturada que resolve um problema real no ecossistema de agentes IA: a descoberta e confiança em skills reutilizáveis. Infraestrutura bem pensada com auditoria, rankings e APIs robustas, ideal para quem constrói sistemas que dependem de múltiplas skills de terceiros.

README

A tradução para o português deste README ainda está na fila.

OpenAgentSkill

OpenAgentSkill is the skill layer for AI agents: helping agents find, compare, and install the right reusable skill automatically.

Think npm for AI Agent Skills: a registry, trust layer, and recommendation API for Codex, Claude Code, Cursor, and other agent runtimes.

Website GitHub Stars License Vercel

Browse Skills | Trending | Audit Reports | API Docs | Submit


What Is OpenAgentSkill?

OpenAgentSkill is a public registry for finding, comparing, auditing, and recommending AI agent skills.

The product is built for three audiences:

AudienceWhat they needOpenAgentSkill surface
Agent buildersFind reliable skills for a taskSearch, rankings, audits, comparisons, agent-friendly APIs
Skill authorsMake their skills discoverable and trustedSubmit flow, claim pages, badges, audit reports
AI agentsQuery skills programmatically/api/agent/skills, /api/agent/recommend, /api/audits/[slug]

The long-term goal is to become the trust and routing layer for agent skills: not just a directory, but the place where humans and agents can decide what is safe, useful, maintained, and worth installing.

Product Surfaces

SurfaceLinkPurpose
Skill directory/skillsSearch and filter the full catalog
Agent skills/agent-skillsSearch entry for reusable agent skills
AI agent skills/ai-agent-skillsCross-agent discovery for Codex, Claude Code, Cursor, and more
Skill clusters/ai-agent-skills/web-scrapingHigh-intent scenario pages for web scraping, coding agents, finance, RAG, World Cup analytics, and more
Skills registry/skills-registryRegistry positioning, trust signals, and install handoffs
Trending/trendingSkills with recent activity signals
Hot/hotHigh-momentum skills
Best lists/bestSEO-ready rankings by use case and category
Audits/auditsSecurity, quality, trust, and adoption-readiness reports
Agent pages/agentsAgent-specific skill discovery
Official creators/officialCreator and organization directories
Comparisons/compareCompetitive and alternative pages
AgentSkills.io alternative/alternatives/agentskills-ioAlternative page for agent skill discovery searches
API docs/api-docsProgrammatic access for agents and apps
Submit/submitSubmit a new skill for review

Key Capabilities

  • High-star GitHub skill indexing with a skill-only search matrix.
  • MCP repositories are intentionally excluded from automated imports.
  • Quality, trust, and audit scoring for each skill.
  • Daily activity aggregates for trending and hot rankings.
  • Agent-friendly search and recommendation APIs.
  • Skill audit pages and embeddable README badges.
  • SEO pages for use cases, alternatives, guides, reports, rankings, and collections.
  • High-intent AI agent skill clusters for concrete workflows such as web scraping, coding agents, finance/quant, RAG, PDF extraction, browser automation, security, and World Cup football analytics.
  • Claimable community-indexed listings with creator/source attribution.
  • Creator-aware X reply drafts for turning public skill discoveries into claimable listings.
  • Manual X Web Intent drafts for compliant social sharing.
  • Optional X OAuth flow for API-based posting when a paid X API plan is available.

Agent API

Recommend Skills For A Task

curl "https://www.openagentskill.com/api/agent/recommend?task=scrape+websites+and+extract+tables&limit=4"

Returns a ranked shortlist with confidence, install command, repository URL, quality profile, decision profile, use cases, and suggested stacks.

Search Skills

curl "https://www.openagentskill.com/api/agent/skills?q=browser+automation&trust=production&limit=5"

Supported filters include q, category, platform, trust, limit, and format=text.

Fetch An Audit

curl "https://www.openagentskill.com/api/audits/crawl4ai"

Returns the stored audit report for a skill, including score, risk level, findings, evidence, and recommendation.

Fetch Rankings

curl "https://www.openagentskill.com/api/agent/rankings"

Returns ranked skills for agent workflows.

README Badges For Skill Authors

Skill authors can add OpenAgentSkill badges to their own repository README.

[![OpenAgentSkill Trust](https://www.openagentskill.com/api/badge/crawl4ai?metric=trust&label=Trust)](https://www.openagentskill.com/skills/crawl4ai)
[![OpenAgentSkill Audit](https://www.openagentskill.com/api/badge/crawl4ai?metric=audit&label=Audit)](https://www.openagentskill.com/skills/crawl4ai/audit)
[![OpenAgentSkill Quality](https://www.openagentskill.com/api/badge/crawl4ai?metric=quality&label=Quality)](https://www.openagentskill.com/skills/crawl4ai)

Replace crawl4ai with the skill slug.

Creator Claim And X Discovery Loop

OpenAgentSkill can index public skills discovered from GitHub, X, creator websites, or community submissions without pretending they are official. Unverified pages are shown as claimable, community-indexed listings until a maintainer verifies ownership.

Useful routes:

# Generate a normal share draft for one listed skill
GET /api/x/share?skill_slug=crawl4ai

# Generate a manual X reply draft for a creator post
GET /api/x/reply-draft?skill_slug=crawl4ai&tweet_url=https://x.com/user/status/123&format=json

The reply draft invites the creator to claim or update the listing. It opens X Web Intent only; OpenAgentSkill does not auto-publish replies.

Auto-Discovery

The indexer scans GitHub for high-signal skill repositories and imports approved matches. It now rotates across scenario-specific query groups so the catalog can expand toward 10,000+ approved skills across developer tools, finance, quant research, documents, data analysis, security, DevOps, RAG, browser automation, commerce, marketing, support, legal, education, productivity, Web3, sports analytics, ML/media, science, and robotics workflows.

Current import rules:

  • Minimum GitHub stars are controlled by INDEXER_MIN_STARS.
  • Per-run target is controlled by INDEXER_RUN_TARGET.
  • Search budget is controlled by INDEXER_MAX_SEARCH_REQUESTS.
  • Search windows rotate across domain query groups each hour, including finance, World Cup/sports analytics, marketing, legal, education, support, productivity, Web3, and other vertical workflows.
  • MCP and Model Context Protocol repositories are excluded from automated imports.
  • Newly imported or updated skill URLs are automatically submitted through the protected IndexNow notification route.
  • A daily baseline IndexNow cron refreshes core discovery pages and the sitemap.
  • Production indexer routes require INDEXER_SECRET.

Useful routes:

# Run the indexer with automation auth
POST /api/indexer/run

# Run a finance/quant-only import
POST /api/indexer/run
{
  "targetNew": 500,
  "minStars": 500,
  "domains": ["finance"],
  "maxSearchRequests": 100
}

# Inspect recent import summaries
GET /api/indexer/logs

# Refresh GitHub star counts
POST /api/indexer/refresh-stars

# Submit fresh URLs to IndexNow with automation auth
POST /api/indexnow/submit
{
  "slugs": ["crawl4ai"],
  "includeBaseline": true
}

Tech Stack

LayerTechnology
FrameworkNext.js 16 App Router
UIReact 19, Tailwind CSS v4, shadcn/ui patterns
DatabaseSupabase Postgres
Auth and privileged writesSupabase SSR plus server-only service role routes
AnalyticsVercel Analytics
DeploymentVercel
AutomationVercel Cron routes and protected API jobs
AI reviewVercel AI SDK / Gateway-compatible review flow

Local Development

git clone https://github.com/Leon-Drq/openagentskill.git
cd openagentskill

pnpm install
cp .env.example .env.local
pnpm dev

Validation:

pnpm run lint
pnpm run build

Environment Variables

VariableRequiredDescription
NEXT_PUBLIC_SUPABASE_URLYesPublic Supabase project URL
NEXT_PUBLIC_SUPABASE_ANON_KEYYesPublic Supabase anon key
SUPABASE_SECRET_KEY / SUPABASE_SERVICE_ROLE_KEYProductionServer-only Supabase key for privileged routes
GITHUB_TOKENRecommendedGitHub API token for higher indexer rate limits
INDEXER_SECRETProductionBearer secret for protected indexer routes
CRON_SECRETProductionBearer secret for scheduled maintenance routes
INDEXER_RUN_TARGETOptionalNumber of new skills to import per run
INDEXER_MIN_STARSOptionalMinimum GitHub stars for bulk imports
INDEXER_MAX_SEARCH_REQUESTSOptionalGitHub search request budget per run
X_CLIENT_IDOptionalX OAuth client ID
X_CLIENT_SECRETOptionalX OAuth client secret
X_ALLOWED_USERNAMEOptionalAllowed X username for token storage

Never commit production secrets. Keep privileged Supabase and X credentials server-only.

Database Setup

Apply SQL files in scripts/ in order. The current schema includes:

  • Skills catalog
  • Profiles and points
  • Activity and feedback events
  • Secure public-write RPCs
  • Indexer run logs
  • X OAuth token storage
  • Claims and skill events
  • Hardened RLS policies
  • Skill audits and daily event aggregates

The latest audit and daily-events migration is:

scripts/013_skill_audits_and_daily_events.sql

Project Structure

app/
  api/
    agent/        Agent-friendly search, rankings, recommendation, feedback
    audits/       Skill audit API
    badge/        SVG badge API
    indexer/      Protected import and maintenance jobs
    x/            X OAuth, Web Intent, and optional posting routes
  skills/         Skill directory and detail pages
  audits/         Audit index
  best/           Best-of ranking pages
  trending/       Trending skills
  hot/            Hot skills
  agents/         Agent-specific pages
  official/       Creator pages
  compare/        Comparison pages
  guides/         Guides and SEO content

lib/
  audits.ts       Audit scoring and normalization
  quality.ts      Quality profiles
  trust.ts        Trust scoring
  decision.ts     Adoption-readiness profile
  rankings.ts     Ranking logic
  indexer/        GitHub discovery and import pipeline
  db/             Supabase data access
  seo/            Programmatic SEO page data

scripts/
  *.sql           Supabase migrations
  *.mjs, *.ts     Content and seed scripts

Roadmap

  • Public skill directory
  • GitHub auto-indexer for high-star skills
  • Scenario coverage matrix for 10,000+ skill growth
  • Skill-only imports with MCP exclusion
  • Quality and trust profiles
  • Audit reports
  • Trending and hot rankings from daily activity
  • README badges
  • Agent recommendation API
  • Programmatic SEO pages
  • Task-based skill evaluations
  • Agent-specific fit scoring for Claude Code, Codex, Cursor, and other agent surfaces
  • Creator claim pages with verified ownership
  • Anonymous install and usage telemetry
  • Semantic search and reranking
  • Public benchmark reports for high-impact skill categories

Contributing

See CONTRIBUTING.md. Useful contribution types include skill submissions, metadata fixes, audit improvements, SEO guide contributions, API improvements, and UI fixes.

License

MIT. See LICENSE.

Ver o original no GitHub

Bom para

Relacionados