Slash-completion of image files should attach them

[mtremsal]

Description

Certain models such as ChatGPT-5 have filesize limits for reading files.

This means that even if the image that I want the model to analyze is inside the codebase, the model cannot call view on it.
This is the case even if I manually /-complete the image's filename when typing the prompt.
As a stop-gap measure, we could ensure that /-completing an image also attaches it.

See example:

[tauraamui]

@mtremsal Just so I understand, how would this stop gap help the too large image size issue. Also, for the stop gap feature you mean, can we have it so selecting (by enter) an image from the file path autocompletion list automatically attaches the image rather than just completing the path?

[mtremsal]

With GPT-5, the model can view attached images, but cannot view images for which it knows the filepath inside the project (this triggers the filesize issue). So any solution that attaches the image to the prompt (without having to manually open the IDE/explorer to drag-and-drop it into Crush) would work as a stop-gap solution. Selecting from the filepath autocompletion menu by pressing enter would work. The only concern might be that such temporary files tend to be in .gitignore'd folders (e.g. /temp), and IIRC the autocomplete ignores them. A more integrated but complex approach would be an MCP or bash integration that somehow views the image directly from puppeteer without persisting it to the disk first.