ollama-vision

本地调用 Ollama qwen3-vl:4b 模型自动压缩并分析图片，支持描述、OCR 文字提取和自定义信息抽取。

This skill appears to do what it claims: compress local images, call a locally running Ollama qwen3-vl:4b model via the local API, and return text results. Before installing or running: 1) Ensure you trust the Ollama build you run locally (the skill posts image data to http://localhost:11434). 2) Be aware that the first use will invoke 'ollama pull qwen3-vl:4b' to download a large model over the network. 3) The included test script reads %USERPROFILE%\.openclaw\media\inbound to find sample images — this is for testing only; review or edit the path if you do not want that directory scanned. 4) The code has a small quality issue: test_skill.py calls analyze_image but the main analysis function is named analyze_image_api (a bug, not malicious). 5) Run check_env.py first and consider running in an isolated environment if you are cautious about large model downloads. If you need higher assurance, verify the Ollama CLI binary and its network behavior independently and review/modify the test script's file-paths before use.

Type: OpenClaw Skill Name: ollama-vision Version: 1.0.0 The ollama-vision skill bundle provides a legitimate tool for local image analysis using the Ollama API and the qwen3-vl:4b model. The implementation in analyze_image.py includes robust features like automatic image compression using Pillow and safe subprocess handling for environment checks. The code communicates exclusively with a local service (localhost:11434) and lacks any indicators of data exfiltration, malicious execution, or prompt injection.