Vision Skill — Development skill for Claude Code
Claude Code 技能:让无原生视觉的模型通过外部多模态 API 看懂图片 Skill that adds vision to models via external multimodal APIs.
How to install Vision Skill
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Autsunset/vision-skill and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Vision Skill does
Claude Code 技能:让无原生视觉的模型通过外部多模态 API 看懂图片 Skill that adds vision to models via external multimodal APIs.
Alternatives in Development
- MCP Builder — Guide for creating high-quality MCP servers to integrate external APIs and services 94.1k ★
- Product Vision — Brainstorm inspiring, achievable product vision statements 7.8k ★
- APIs Version Update — API Version Update Guide for Hydrogen 2.1k ★
README
vision-skill
一个 [Claude Code](https://docs.claude.com/en/docs/claude-code) 技能(Skill),让**没有原生视觉能力的模型**也能通过外部多模态 API「看懂」图片:截图、照片、图表、示意图、OCR 文字识别等都能分析。
兼容 OpenAI、Anthropic,以及任意 OpenAI / Anthropic 兼容的视觉端点(Azure OpenAI、OpenRouter、本地多模态模型、第三方中转 API 等)。
这是一个**后备视觉方案**:如果当前模型或会话已经能直接查看图片,应优先使用原生视觉;只有原生视觉不可用或确实失败时,才调用此外部 API,避免额外延迟、费用和图片传输。
功能
- 内置能力检查:原生视觉可用时不调用外部 API
- 读取本地图片(PNG / JPG / JPEG / GIF / WebP / BMP / TIFF),也支持
data:image/...;base64,...URI - 自动 base64 编码后调用视觉模型
- 同时支持 OpenAI 与 Anthropic 两种 API 格式
- 自定义分析 prompt(不传则给出详细描述)
- 修复了 Windows 下中文输出的乱码问题(强制 UTF-8)
安装
把 `SKILL.md` 与 `scripts/` 放进 Claude Code 的 skills 目录:
~/.claude/skills/vision-skill/
├── SKILL.md
└── scripts/
├── vision-skill.sh
└── vision.py
一键安装:
mkdir -p ~/.claude/skills/vision-skill
cp -r SKILL.md scripts ~/.claude/skills/vision-skill/
cp .env.example ~/.claude/skills/vision-skill/.env
# 编辑 .env 填入你的配置
配置
三种方式,优先级:**环境变量 > .env 文件 > 配置文件**
方式一:.env 文件(推荐)
在 skill 目录创建 `.env` 文件:
cp .env.example ~/.claude/skills/vision-skill/.env
# 编辑 .env,填入真实配置
VISION_API_KEY=sk-your-key-here
VISION_API_BASE=https://api.openai.com/v1
VISION_MODEL=gpt-4o
VISION_API_FORMAT=openai
VISION_MAX_TOKENS=4096
方式二:配置文件
复制示例并填入你的密钥:
cp vision-config.example.json ~/.claude/vision-config.json
# 编辑 ~/.claude/vision-config.json,填入真实 api_key
{
"api_key": "sk-your-key-here",
"api_base": "https://api.openai.com/v1",
"model": "gpt-4o",
"api_format": "openai",
"max_tokens": 4096
}
方式三:环境变量
| 变量 | 默认值 | 说明 |
|---|---|---|
VISION_API_KEY |
(必填) | API 密钥 |
VISION_API_BASE |
https://api.openai.com/v1 |
API 地址 |
VISION_MODEL |
gpt-4o |
模型名 |
VISION_API_FORMAT |
openai |
openai 或 anthropic |
VISION_MAX_TOKENS |
4096 |
返回最大 token 数 |
环境变量优先级最高,适合 CI/CD 或临时覆盖。
Anthropic 格式
设 `VISION_API_FORMAT=anthropic`,默认走 `https://api.anthropic.com/v1/messages`,用 `x-api-key` 头与 `ant
Related Skills
Cc Vision Hook
Claude Code hook toolkit that gives vision-blind models a text description of pasted/tool-produced images.
Claude API Skill
Professional-grade Claude API skill framework. Comprehensive guide covering Messages API, streaming, tool use,
Dexscreener Unofficial CLI + MCP + Skills
You are a token scanning specialist using the Dexscreener Unofficial CLI (not affiliated with or endorsed by D
Public APIs Live
Every public API in one place — aggregated, deduped, and auto-checked for reachability. Refreshed daily. npm +
Lathe
Agentic-friendly CLI generator for APIs: turn Swagger, OpenAPI, and google.api.http protos into single-binary
Laravel API Skill
A comprehensive Claude skill for building production-grade Laravel REST APIs with clean architecture, type saf
Related Agents
OCR Reader
Batch OCR re-reader for low-confidence book pages. Reads page screenshot images via multimodal Read tool and e
External API Researcher
Researches an external API by fetching and analyzing its documentation. Extracts authentication methods, endpo
Be FastAPI Expert
Expert FastAPI developer for building high-performance async Python APIs. Use for FastAPI projects, Python asy