CHNAICORP

Multimodal Vision Skill

AI community

SOURCE https://github.com/CHNAICORP/multimodal-vision-skill

Description

One image-recognition engine for any AI agent — 3 protocols (OpenAI Chat/Responses, Anthropic Messages), multi-provider (DeepSeek, GLM + any compatible endpoint). Works with Claude Code, Codex, Cursor, ZCode, OpenClaw & more.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

multimodal-vision

**English** | [简体中文](README.zh-CN.md)

**One image-recognition engine for any AI agent** — three wire protocols, multiple providers, a single CLI.

Give your agent the ability to *see*. `multimodal-vision` lets Claude Code, Codex, Cursor, ZCode, OpenClaw, Hermes — or any IDE/ADE with terminal access — analyze, describe, compare, and OCR images through hosted vision LLMs, without writing any glue code.

your agent ──▶ vision.py analyze img.png ──▶ [ openai-chat │ openai-responses │ anthropic ] ──▶ DeepSeek · GLM · any compatible endpoint

Highlights


Three protocols, one CLI	OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages. Switch with `--protocol`; the engine handles each wire format (data-URI vs `input_image` vs base64 source blocks) for you.
Provider registry	Presets for DeepSeek (`deepseek-v4-flash-vision-exp`) and Zhipu GLM (`glm-5.3-flash`); add any OpenAI- or Anthropic-compatible endpoint by editing one JSON file.
Multi-image & URL input	Mix local paths and remote URLs in one call; everything is normalized to base64 so all protocols behave identically.
Reasoning-model aware	Handles `reasoning_content` / `thinking` blocks; warns when the token budget is too small to produce a final answer.
Operational hygiene	Unique-prefix model matching (`-m glm`), exponential-backoff retries on 429/5xx, live self-check, `--json` output, magic-byte image sniffing.
Minimal dependencies	Python 3.8+ and `requests`. No Pillow, no framework.

Quick start

git clone https://github.com/CHNAICORP/multimodal-vision-skill.git ~/.agents/skills/multimodal-vision
cd ~/.agents/skills/multimodal-vision
cp config/providers.example.json config/providers.json   # then fill in your API keys
python3 scripts/vision.py check                          # live self-test

Analyze your first image:

python3 scripts/vision.py analyze photo.jpg -q "Descri