# Vision Reader

> >

- **Type:** Skill
- **Install:** `agentstack add skill-liu-zhangzhu-opencode-vision-skill-opencode-vision-skill`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [Liu-Zhangzhu](https://agentstack.voostack.com/s/liu-zhangzhu)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [Liu-Zhangzhu](https://github.com/Liu-Zhangzhu)
- **Source:** https://github.com/Liu-Zhangzhu/opencode-vision-skill

## Install

```sh
agentstack add skill-liu-zhangzhu-opencode-vision-skill-opencode-vision-skill
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Vision Reader Skill

Enables pure-text LLMs (DeepSeek, etc.) to read binary files by calling Python tools.

## Install Path

All tools are located at `~/opencode-vision-skill/tools/`. The `~` means your home directory:
- **Windows**: `%USERPROFILE%\opencode-vision-skill\tools\`
- **macOS/Linux**: `~/opencode-vision-skill/tools/`

If you installed elsewhere, adjust the paths below accordingly.

## Available Tools

| Tool | Command |
|------|---------|
| `pdf-reader` | `python ~/opencode-vision-skill/tools/pdf-reader.py ` |
| `image-reader` | `python ~/opencode-vision-skill/tools/image-reader.py ` |
| `ppt-reader` | `python ~/opencode-vision-skill/tools/ppt-reader.py ` |
| `screenshot` | `python ~/opencode-vision-skill/tools/screenshot.py [x,y,w,h]` |

Always use **absolute file paths** for the filepath argument.

## Tool Details

### pdf-reader — Read PDF files
```
python ~/opencode-vision-skill/tools/pdf-reader.py "/absolute/path/to/file.pdf"
```
Returns Markdown with page headings, text content, and tables.

### image-reader — Read and describe images
```
python ~/opencode-vision-skill/tools/image-reader.py "/absolute/path/to/image.png"
```
Returns three sections:
- **Metadata** — Format, mode, size, DPI
- **Visual Description** — AI-generated caption describing objects, layout, arrows, diagrams
- **OCR Text** — Recognized text from the image

### ppt-reader — Read PowerPoint files
```
python ~/opencode-vision-skill/tools/ppt-reader.py "/absolute/path/to/slides.pptx"
```
Returns slide-by-slide text with titles, bullet points, and tables.

### screenshot — Capture the screen
```
python ~/opencode-vision-skill/tools/screenshot.py
```
Captures full screen. For a region: `python ~/opencode-vision-skill/tools/screenshot.py 0,0,800,600`.
Returns the saved PNG file path.

## Notes

- The first `image-reader` call downloads Florence-2 model (~400MB) if not pre-downloaded
- Tesseract OCR is optional — image-reader works without it using easyocr
- Screenshot requires a display server (X11/Wayland on Linux, native on Windows/macOS)

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [Liu-Zhangzhu](https://github.com/Liu-Zhangzhu)
- **Source:** [Liu-Zhangzhu/opencode-vision-skill](https://github.com/Liu-Zhangzhu/opencode-vision-skill)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-liu-zhangzhu-opencode-vision-skill-opencode-vision-skill
- Seller: https://agentstack.voostack.com/s/liu-zhangzhu
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.