| name | image |
| description | Work with image files safely; do not decode them as text, use image URLs for referencing, and use Gemini only for generation or editing requests. |
| tools | ["gemini_image","get_image_url","request_clarification"] |
image
Overview
Use this skill for image files such as .png, .jpg, .jpeg, .gif, .bmp, .webp, and .svg.
Rules
- Never call
read_file on a binary image.
- Use
get_image_url when the user needs to reference an uploaded workspace image in Markdown, HTML, or chat.
- Use
gemini_image only when the user wants image generation or image editing.
- Do not claim to have visually inspected an image if the runtime only has file metadata and no vision input path for that file.
Workflow
-
Determine the request type.
- Reference an existing workspace image
- Generate a new image
- Edit an existing image
- Describe or inspect image content
-
Reference flow.
- If the task is to include or share an existing workspace image, call
get_image_url.
- Use the returned public URL in the output artifact or response.
-
Generation or editing flow.
- If the user is asking for creation or editing in substance, use
gemini_image.
- Keep the prompt grounded in the user's stated goal.
-
Visual-inspection limitation handling.
- If the user asks what is inside a workspace image and there is no supported vision-read path for that file in the current runtime, say so plainly.
- Ask for either:
- a direct description request using a supported image input path
- a screenshot embedded in a PDF or document that can be queried another way
- the specific text/context they want extracted
Good uses
- Insert or share uploaded charts and diagrams with public URLs
- Generate a new illustration, concept image, or mockup
- Edit or restyle an image with Gemini
Avoid
- Treating PNG or JPG bytes as text
- Pretending an image was visually analyzed when it was not