convez.ai
SkillIntegração & APIsIntermediário
outstandingcandy

Claude Skill: Análise de Imagem com Gemini

outstandingcandy/claude-skill-gemini-image

Integra Gemini Pro para análise visual avançada no Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill para Claude Code que delega análise de imagens ao modelo Gemini 3 Pro Preview via CLI. Útil quando você precisa de uma segunda opinião especializada em OCR denso, leitura de diagramas ou comparação cross-model de conteúdo visual.

Instalação

$npx convez-ai outstandingcandy/claude-skill-gemini-image

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

  • OCR pesado e extração de texto em imagens complexas
  • Análise detalhada de diagramas e esquemas técnicos
  • Validação cross-model de conteúdo visual
  • Processamento de screenshots e capturas de tela de alta densidade

Veredito

Solução prática que estende capacidades visuais do Claude delegando tarefas específicas ao Gemini Pro. Requer setup prévio (gemini-cli + autenticação Google) e é mais adequada para usuários que já dominam Claude Code e precisam de análise visual multi-modelo.

README

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

claude-skill-gemini-image

A Claude Code skill that delegates image analysis to Google's gemini CLI using the gemini-3-pro-preview model.

Useful when you want a second opinion from Gemini Pro on dense OCR, fine-grained scene description, or sports-video tracklet thumbnails — anywhere Claude's own multimodal vision is good but you want Pro-tier Gemini specifically.

Why a skill (not just Read an image)

Claude Code is already multimodal. Reach for this skill when:

  • The user explicitly asks for gemini analysis.
  • The task is heavy OCR / dense diagram reading where Gemini Pro is meaningfully stronger.
  • You want a cross-model check on what's in an image.

Prerequisites

sudo npm install -g @google/gemini-cli
gemini   # first run: log in with your Google account (free tier is fine)

Verify with command -v gemini and gemini --version.

Install

Drop SKILL.md into your user-level skills directory:

mkdir -p ~/.claude/skills/analyze-image
curl -fsSL \
  https://raw.githubusercontent.com/outstandingcandy/claude-skill-gemini-image/main/SKILL.md \
  -o ~/.claude/skills/analyze-image/SKILL.md

Restart your Claude Code session so the skill list reloads.

Usage

In Claude Code, just ask:

  • "Analyze /tmp/foo.png for me"
  • "OCR this screenshot at /abs/path/to/img.jpg"
  • "Use gemini to tell me what's in https://example.com/x.jpg"

Claude will run a preflight to confirm gemini is installed, copy the image to a workspace gemini-cli can read, invoke gemini -m gemini-3-pro-preview -p "..." @<path>, and relay the answer with a footer disclosing which model produced it.

What this skill works around

gemini-cli enforces two strict rules that bite when you point it at images in real projects:

  1. .gitignore is enforced as a read-block. Files under any ignored directory (output/, build/, etc.) get refused with File path '...' is ignored by configured ignore patterns. even when explicitly referenced via @.
  2. Only the cwd subtree is the workspace. Absolute paths outside cwd are refused with Path not in workspace.

The skill solves both by copying the target image into /tmp and running gemini-cli from /tmp with --skip-trust.

License

MIT

Ver o original no GitHub

Bom para

Relacionados