convez.ai
SkillAutomaçãoAvançado
thesved

Epic Skills

thesved/epic-skills

Multi-modelo (Opus+Gemini+Codex) com thinking discipline e utilitários sharp

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto curado de skills para Claude Code com suporte a múltiplos modelos (Opus, Gemini, Codex em paralelo), disciplina de raciocínio aprofundado e utilitários práticos. Permite delegar tarefas ao modelo mais adequado e comparar respostas divergentes como sinais de complexidade.

Instalação

$npx convez-ai thesved/epic-skills

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

Casos de uso

  • Encadear raciocínio aprofundado com consultas paralelas a múltiplos modelos
  • Automação de código com verificação in-browser via Chrome DevTools Protocol
  • Processamento de vídeo, tradução e síntese de voz via Gemini

Veredito

Ferramental sofisticado para quem domina Claude Code e quer orquestrar múltiplos modelos com eficiência. Exige setup cuidadoso (gemini-cli, codex, chaves de API), mas oferece workflow potente com economia de tokens via modo caveman e raciocínio delegado ao modelo certo conforme o problema.

README

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

epic-skills

A curated set of Claude Code skills: a multi-model board (Opus + Gemini + Codex), the thinking discipline to use it well, and a few sharp utilities. The seats let Claude delegate to whichever model is actually strongest - Gemini for video/translation/TTS, Codex for spec-grade prompts - and weigh their disagreements.

Install - one line, any agent

Paste this to your coding agent (Claude Code, etc.):

Clone https://github.com/thesved/epic-skills, run bash epic-skills/.claude/skills/epic-install/install.sh copy --target ~/.claude/skills, then walk me through the epic-install skill (API keys, CLIs, verification).

Or do it yourself:

git clone https://github.com/thesved/epic-skills ~/epic-skills
bash ~/epic-skills/.claude/skills/epic-install/install.sh copy --target ~/.claude/skills
# then, in Claude Code:  /epic-install   →  sets up keys, CLIs, and verifies the board

Install to a project instead of globally by passing --target <project>/.claude/skills. Update later with /epic-update (preserves any edits you made).

How it fits together - one workflow

The skills are designed to chain. A typical session:

/orient        →  where am I, what's next (caveman-terse map of the project)
/think  <task> →  Opus reasons hard, escalates to the board when it matters
/board  <q>    →  Opus + Gemini + Codex answer in parallel; disagreement = signal
   ├─ /gemini-bridge  →  video/YouTube, non-English copy, image, TTS, long docs
   └─ /codex-bridge   →  spec-grade prompts for Opus, gpt-image, a 2nd-architecture take
/grill-me      →  get interviewed until the plan has no soft spots
/ask           →  (the inverse) Claude asks YOU only when genuinely blocked
/chrome        →  drive a real Chrome over CDP to verify/QA the change in-browser
/wrap          →  close the session so a cold start resumes with zero context loss

Plus two utilities: /caveman (terse mode, ~75% fewer tokens) and /skill (create, change, review, or doctor a Claude Code skill: the meta-tool that built and maintains this set).

Prerequisites

ToolForInstall
gemini CLIGemini seat + medianpm i -g @google/gemini-cli
codex CLICodex seatnpm i -g @openai/codex → codex login
jq, curlGemini seat plumbingyour package manager
ffmpeg (opt)TTS/audio outputyour package manager

API keys (set up by /epic-install, stored in your OS secret store - Keychain / libsecret / pass):

KeyNeeded forGet it at
GEMINI_API_KEYGemini seat, video/TTS/imagehttps://aistudio.google.com/apikey
OPENAI_API_KEYoptional - Codex image-gen / API modelshttps://platform.openai.com/api-keys
OPENROUTER_API_KEYoptional - throttle fallbackhttps://openrouter.ai/keys

_model-cache

Model ids, pricing, and call-shapes live in _model-cache/. They go stale - providers hot-swap aliases and prices. Refresh with bash _model-cache/update.sh and verify with bash _model-cache/verify.sh --cheap. Never hardcode model ids; the cache + self-updating aliases are the pin.

License

MIT - see LICENSE.

Ver o original no GitHub

Bom para

Relacionados