convez.ai
SkillAutomaçãoIntermediário
yeb07011

Agent Skill Studio

yeb07011/agent-skill-studio

Gera e valida skills de IA para múltiplos agentes com linting de segurança

Ver no GitHub
InicialPluginScore · de 0 a 100

Ferramenta para gerar, validar e exportar skills reutilizáveis para agentes de IA (Claude Code, Codex, Cursor, Gemini CLI). Converte prompts e SOPs em pacotes de skills portáveis com linting de segurança integrado, detectando riscos como comandos perigosos, vazamento de segredos e execução autônoma insegura.

Instalação

$npx convez-ai yeb07011/agent-skill-studio

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
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Criar skills padronizadas para equipes de desenvolvimento que usam múltiplos agentes de IA
  • Validar segurança de instruções antes de compartilhar skills em repositórios
  • Converter SOPs e workflows em pacotes portáveis entre Cursor, Claude Code, Codex e Gemini

Veredito

Solução prática para padronizar skills de IA com foco em segurança. CLI + Web UI local tornam acessível tanto para developers quanto para não-técnicos, embora o ecossistema multi-agente ainda esteja em evolução.

README

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

Agent Skill Studio

npm License: MIT Build Stars TypeScript

Turn prompts, SOPs, and workflow notes into reusable AI agent skills for Claude Code, Codex, Cursor, and Gemini CLI.

Agent Skill Studio gives AI coding tools a portable skill package format: generate tool-specific instructions, lint them for safety and completeness, and export a manifest that can be shared in a repo or future registry.

Web UI preview

Why Star It

  • One prompt, four targets: Codex, Claude Code, Cursor, and Gemini CLI.
  • Generate / lint / export: Markdown skill files plus agent-skill.json.
  • Safe skill linting: catches vague goals, missing review steps, dangerous commands, network risk, deletion risk, and secret leakage.
  • Local-first MVP: no LLM API key required; deterministic templates today, provider interface ready for later.
  • CLI + Web UI: scriptable for maintainers, friendly for non-experts.

Quick Start

pnpm install
pnpm test
pnpm build
pnpm agent-skill-studio generate --name frontend-taste --description "Make landing pages visually polished"
pnpm agent-skill-studio lint ./generated/.claude/skills/frontend-taste/SKILL.md
pnpm dev

Example Output

generated/
  AGENTS.md
  agent-skill.json
  .codex/skills/frontend-taste/SKILL.md
  .claude/skills/frontend-taste/SKILL.md
  .cursor/rules/frontend-taste.mdc
  .gemini/frontend-taste.md
  examples/prompt.md
  README.md

CLI

pnpm agent-skill-studio templates
pnpm agent-skill-studio generate --template code-review
pnpm agent-skill-studio generate --name frontend-taste --description "Make landing pages visually polished"
pnpm agent-skill-studio lint ./generated/.claude/skills/frontend-taste/SKILL.md

Generated packages include target files for Claude Code, Codex, Cursor, Gemini CLI, a generic AGENTS.md, example prompt, README, and export manifest.

Web UI

pnpm dev

The Web UI runs locally in the browser. It includes skill name and description inputs, template cards, Generate and Lint actions, lint score card, generated file preview, and copy button.

Safe Skill Linting

The linter returns:

{
  "score": 100,
  "level": "excellent",
  "issues": [
    {
      "severity": "info",
      "message": "No major structure or safety issues found.",
      "suggestion": "Review the skill against your actual workflow before sharing it."
    }
  ]
}

It checks for missing structure, missing success criteria, missing human review, vague objectives, unsafe autonomous execution, network command risk, file deletion risk, dangerous commands, and secret exfiltration risks such as .env, GITHUB_TOKEN, OPENAI_API_KEY, and private keys.

Built-In Templates

TemplatePurpose
frontend-tasteHelp AI agents generate more polished frontend pages.
code-reviewReview code for bugs, regressions, security risks, and missing tests.
product-manager-interviewPrepare for AI product manager interviews.
ai-evaluationDesign LLM evaluation tasks, rubrics, and reports.
github-trending-researchResearch GitHub trending projects and produce reports.
xiaohongshu-content-agentPlan, generate, and review Xiaohongshu content.

See examples/gallery.md for the full gallery.

Screenshots

Web UICLILint report
Web UI previewCLI previewLint report preview

Docs

Project Structure

apps/web              Vite + React local UI
packages/core         Shared generator, linter, templates, formatters
packages/cli          Node.js CLI entrypoint
examples              Source prompts, SOPs, and template gallery
generated             Default output directory

Contributing

Contributions are welcome. Good first areas: add templates, improve formatters, add lint rules, polish the Web UI, or contribute real agent workflow examples.

Before opening a PR:

pnpm lint
pnpm typecheck
pnpm test
pnpm build

Safety Notes

Generated skills should be reviewed before use. Do not put secrets, credentials, API keys, private keys, private repository data, private user data, or environment files into a skill.

Agent Skill Studio is a generator and linter, not a sandbox. The user and the target AI coding tool are still responsible for reviewing and approving actions.

Ver o original no GitHub

Bom para

Relacionados