Annotate pasted images (arrows, circles, numbered comments) before sending

[arthureffting]

Problem

When I paste a screenshot into the chat input, the only thing I can do with the thumbnail is look at it or remove it. Most screenshots I send are about one specific spot ("this button is misaligned", "this error here"), and I end up describing positions in words, or annotating in another tool first and pasting the result.

Proposal

Clicking an attached image thumbnail opens a lightweight annotation editor:

• Tools: line, arrow, circle, and numbered marker.
• Comments: each numbered marker can carry a short comment. Undo and delete for every mark.
• Done returns to the composer with the annotated image attached.

What the agent receives

• The image with the marks flattened into it, so any vision model sees where they are.
• A numbered list appended as text, for example:
  1. This button is misaligned.
  2. This error appears after saving.

Comments travel as text rather than drawn pixels, because small rendered text is read unreliably by models, and the numbers on the image tie each comment to its spot. The original image should stay recoverable (re-open the thumbnail to edit marks again).

Out of scope for a first version

Freehand drawing, rectangles, text boxes and cropping. They can follow if the basics prove useful.

Why

It turns "the third button in the second row, the one next to the dropdown" into a circle and a number, and it makes UI feedback loops with the agent much faster and less ambiguous.