agent-vision-toolkit
A vision toolkit and skill that gives a text-only coding agent, such as one running DeepSeek, image Q&A, long-screenshot OCR, frontend UI restoration and GUI automation.
agent-vision-toolkit gives a text-only large language model the ability to work with images. It provides a set of vision-tool command-line programs, plus a skill file that teaches an agent when to use each one, so any agent that can invoke a shell can call them for image question answering, long-screenshot OCR, restoring a frontend UI from a screenshot, and GUI automation.
An optional integration layer adds a transparent local proxy and single-file native plugins so pasted images and an agent's own built-in image tools both work without extra tool installation or prompting. The project reports that it has been verified end to end in Codex with DeepSeek, and in Claude Code, Pi, Oh My Pi and OpenCode.
It is written in Python and released under the MIT license.