#400400th of 455 in Skills, plugins and prompts
dsh-vision-toolkit
Vision tools and a Skill for text-only DeepSeek Harness agents
- Stars
- 918
- License
- MIT
- Last commit
- Oct 2026
- Last release
- Oct 2026
- Language
- TypeScript
Missed the latest GitHub check; these facts may be older.
Overview
A plugin for DeepSeek Harness (DSH) that adds 10 vision tools (image Q&A, grounding with pixel coordinates, crop, SVG trace, pixel diff, long-screenshot OCR) and a bundled `vision-skills` Skill. Image understanding goes to a vision model you configure; crop, trace and diff run locally in an isolated Python runtime. Pasting an image in DSH Web routes a text-only model to a `(Vision Toolkit)` variant.
Who it is for: DeepSeek Harness users who want text-only models to handle images
Strengths
- Ten tools: OCR, grounding, crop, SVG trace, pixel diff, HTML screenshots
- Skill ships five playbooks, including UI restoration and long-screenshot OCR
- Vision provider can use Chat Completions, Responses or Anthropic Messages
- Coordinates are original-image pixels, so output feeds crop or trace directly
Weaknesses
- Works only inside DeepSeek Harness, not other agent hosts
- Needs an external vision model API key for image understanding
- First run downloads Python (about 35 MB) and dependencies
- DSH Desktop 2.0.1 marketplace install has known issues; terminal install needed
What it needs
- no GPU
- Needs DeepSeek Harness (dsh CLI), Python 3.11+ (or auto-downloaded standalone Python), Pillow, NumPy, vtracer, External vision model API
- Models: OpenAI Chat Completions-compatible, OpenAI Responses, Anthropic Messages, Gemini 3.7 Flash (default)
Also in Skills, plugins and prompts
See all 455| Rank | Project | Score |
|---|---|---|
| 1 | claude-memPersistent cross-session memory plugin for Claude Code and other coding agents | 97 out of 100 |
| 2 | impeccableDesign skill with 24 slash commands for AI coding agents' frontend output | 97 out of 100 |
| 3 | gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles | 96 out of 100 |
| #4 | rufloAgent harness adding swarms, memory and hooks to Claude Code and Codex | 96 out of 100 |
| #5 | cavemanShortens coding-agent replies and compresses tool output before the model reads it | 95 out of 100 |