vision

Multi-provider vision tool. Feed it a prompt + image path, get back a text description of the image. Bridges the gap for text-only base models (e.g. DeepSeek) that cannot see images natively, and gives multimodal models a cheaper/faster alternate vision path.

Quick start

uv run "${CLAUDE_SKILL_DIR}/scripts/vision.py" [--provider <name>] <image_path> <prompt>

uv run reads the PEP 723 dependency block at the top of vision.py and auto-installs openai into an ephemeral environment on first run — no manual install step, no global pollution. Prerequisite: have uv on the PATH (pip install uv or winget install astral-sh.uv on Windows).

When --provider is omitted, the provider resolves by: --provider flag → VISION_PROVIDER env → first API key found in ~/.env (by file line order) → doubao.

The same uv run command works in Git Bash and PowerShell on Windows, and on macOS/Linux.

First-time setup

A ready-to-edit template ships at .env.example (covers every provider, endpoint, model, and tuning var this skill reads). Copy it to ~/.env and fill in your keys: