When a plain screenshot is enough
If the screen has one obvious broken element and no look-alikes, a clean crop plus one sentence often works. Do not over-tool simple tasks.
Examples: a single missing icon, a full-page blank state, or a crash dialog with unique copy. In those cases freeform capture tools are fine.
When numbers win
Numbers win when multiple regions matter, when elements look similar, or when the relationship between regions is the bug ("[2] should align with [1]"). Freeform arrows are fine for humans; models benefit from stable labels that appear in both the image and the text.
Arrows and freehand scribbles also compress poorly in vision: color, thickness, and placement vary. Consistent numbered badges are easier for a model to bind to tokens like [2].
How Casso uses numbers
Each box is numbered in draw order. Notes attach to those indices. The prompt references [n] while the annotated composite shows matching badges. Crops give the agent a closer look without losing the overview.
That pairing - overview for layout, crop for detail, text for intent - is what turns a screenshot into a work order.
Quick decision table
Use this when you are unsure whether to mark:
| Situation | Prefer |
|---|---|
| One unique broken control | Plain crop + one sentence |
| Many look-alike buttons | Numbered boxes |
| Spacing between two regions | Two numbers + relation note |
| Human-only design review | Freeform arrows OK |
| Agent must act without a meeting | Numbers + structured notes |
Related
Try Casso free for 7 days
Numbered screen annotations for Claude Code, Codex, and other agents. 7-day free trial, then $29 once - not a subscription.