Agent Vision Toolkit — vision-skills

Give text-only coding agents a practical visual workflow with the original vision-skills Skill from Anionex/agent-vision-toolkit. This is the general CLI-based edition.

The package teaches an agent which tool to use, how to combine visual evidence, and how to verify a result. It includes:

• SKILL.md and agent interface metadata.
• Five playbooks for long-screenshot OCR, screenshot-to-UI reconstruction, graphic extraction/restoration, structured diagrams, and screenshot-guided GUI work.
• Five Python helpers for long-screenshot OCR, foreground extraction, HTML screenshots, dominant colors, and pixel comparison.

The workflow combines glance for image questions and OCR, ground for locating a target, detect for element inventories, trace for SVG geometry, and crop for extracting image regions.

Requirements: a coding agent that can run shell commands, Python 3.11+, and the agent-vision-toolkit CLI tools installed separately and available on PATH. Configure your own vision API for image Q&A, grounding, detection, and OCR. Pillow and NumPy are needed for relevant image-processing tools; vtracer is optional for the outline fallback. HTML screenshots require Chrome, Chromium, or Edge.

This ZIP contains the published Skill and bundled helpers, not the repository-level CLI runtime, proxy, native extensions, or API credentials. The optional seamless-integration layer is not required for CLI-based use.

Free and MIT-licensed. Any external vision API charges depend on your own provider. Packaged from public repository commit 328c0ae, with the original Skill files unchanged and the MIT license included.

Source and setup: https://github.com/Anionex/agent-vision-toolkit