dsh-vision-toolkit: A Vision Toolkit Plugin for DeepSeek Harness
anionex/dsh-vision-toolkit
A DeepSeek Harness plugin that gives text-only agents vision: intent-aware image Q&A, long-screenshot OCR, UI restoration, and pixel-level verification.
Give text-only models the ability to see images, OCR, and restore UI, integrating vision into DeepSeek Harness agent workflows.
dsh plugin --profile web add github:anionex/dsh-vision-toolkit- Category
- Tools & Capabilities
- Platform
- DSH-Plugin
- Author
- anionex
- Distribution
- Plugin
dsh-vision-toolkit Key Features
dsh-vision-toolkit Repository
dsh-vision-toolkit Repo Summary
DSH Vision Toolkit is a vision-tool plugin for DeepSeek Harness (DSH), maintained by Anionex under the MIT license. It brings agent-vision-toolkit into DSH as a native Profile Bundle, enabling text-only models to handle visual tasks. The plugin solves the problem of pure-text LLMs being unable to see images, perform OCR, or restore UI. Core capabilities include intent-aware image Q&A, long-screenshot OCR, UI restoration, pixel-level grounding, pixel diff, managed Artifacts, and a Web UI, with ten independent tools replacing shell glue with structured schemas and Agent-scoped progressive exposure.
What core vision capabilities does the DSH Vision Toolkit DSH plugin support?
The plugin provides ten vision tools covering intent-aware image Q&A, long-screenshot OCR, UI restoration (screenshot to editable HTML/CSS), pixel-level grounding, and pixel verification (diff). It emphasizes "looking with intent": the method carries the reason for looking, moves from the whole image to targeted regions, and verifies coordinates, colors, geometry, and differences with focused tools instead of accepting generic descriptions as evidence. These tools are loaded as a versioned Skill and only expose the relevant schemas when the current task needs them, avoiding waste.
How do I install DSH Vision Toolkit into DeepSeek Harness?
Use the following command for the Web profile:
dsh plugin --profile web add github:anionex/dsh-vision-toolkit
After installation, the plugin manages credentials via DSH Credentials, handles runtime preparation through lifecycle management, logs results as structured Session-log entries, and provides previewable Artifacts and dedicated Web cards.
Which DeepSeek Harness versions and platforms does DSH Vision Toolkit support?
The plugin targets both Web and Headless DSH profiles, requiring Node.js ^22.19 or >=24 and Python 3.11+. It deeply integrates with the DSH ecosystem, depending on official packages like @deepseek-ai/dsh-agent, @deepseek-ai/dsh-tools, and @deepseek-ai/dsh-skill, ensuring seamless cooperation with DSH's agent, tool, skill, settings, and session modules.
What is the verification and update status of DSH Vision Toolkit?
The project is verified by 168 tests, with the current release at v0.1.8, last updated on 2026-08-16. It is maintained by Anionex, with the project website at agent-vision.anionex.me and the upstream toolkit at agent-vision-toolkit. The plugin delivers the committed P0 and P1 product scope; the P2 ctx.visionToolkit service remains unpublished until an independent plugin becomes a real consumer.
What are typical use cases for DSH Vision Toolkit?
The official documentation shows two upstream reference cases: infographic restoration (screenshot to editable HTML/CSS) and screenshot-guided debugging. In a live DSH Web session, image Q&A and screenshot debugging workflows are demonstrated. These scenarios illustrate how the plugin integrates vision capabilities into Agent workflows, allowing text-only models to handle vision-intensive tasks.