Epic Skills
thesved/epic-skills
Multi-modelo (Opus+Gemini+Codex) com thinking discipline e utilitários sharp
Ver no GitHubConjunto 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-skills1 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
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, runbash epic-skills/.claude/skills/epic-install/install.sh copy --target ~/.claude/skills, then walk me through theepic-installskill (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
| Tool | For | Install |
|---|---|---|
gemini CLI | Gemini seat + media | npm i -g @google/gemini-cli |
codex CLI | Codex seat | npm i -g @openai/codex → codex login |
jq, curl | Gemini seat plumbing | your package manager |
ffmpeg (opt) | TTS/audio output | your package manager |
API keys (set up by /epic-install, stored in your OS secret store - Keychain / libsecret / pass):
| Key | Needed for | Get it at |
|---|---|---|
GEMINI_API_KEY | Gemini seat, video/TTS/image | https://aistudio.google.com/apikey |
OPENAI_API_KEY | optional - Codex image-gen / API models | https://platform.openai.com/api-keys |
OPENROUTER_API_KEY | optional - throttle fallback | https://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.