image_to_text

Files (4)

• SKILL.md
• __init__.py
• __init__.py · scripts/
• core.py · scripts/

SKILL.md (first lines)

**First check whether you already have the image.** When the conversation carries the image as an attachment — a vision-capable model receiving an upload, or a tool result that returned the image itself — you can SEE it: describe what you see and never run this script for it. This skill exists for what you cannot see directly: a bare file path or URL with no attached image, and a model without native vision.

**Recognize an image (local file path OR URL):**

The `image_path` argument accepts BOTH a local absolute file path AND an http/https URL. If the user uploaded an image, the middleware message tells you its exact location — pass that value directly to `image_path`.

Use the **`terminal` tool** to run the skill script (the `python_repl` tool cannot import skill modules because its builtins are restricted). Run:

```bash
python -c "from skills.builtin.core.image_to_text.scripts import itt; print(itt(image_path='{replace with the image location}', user_text='{replace with the user question about the image}'))"
```

- `image_path`: the image's local path or URL given in the middleware message.
- `user_text`: the user's question/instruction about the image.

**Priority:** When the user uploads or asks you to look at an image, you MUST run this skill script and actually recognize the image — do not answer with a generic response without looking at it. The printed output is the ground-truth recognition result.

From lasertsole/sherry_agent, MIT.

More from lasertsole/sherry_agent