Browser
Picking elements
Select an element on the page and hand the agent the exact one, instead of writing a paragraph describing it.
Describing an element is hard (“the grey button on the left” is rarely unique). Picking is pointing instead of describing.
How to use it
- Open the target page in Chrome.
- Open the DSH side panel and enter picking mode.
- Move the cursor over the page; candidate elements highlight.
- Click to select — the element’s details enter the conversation as context.
- Then say what you want done with that element.
What gets carried over
What enters the context is not just “a button” but its DOM structure, its visible text and its key attributes — enough for the agent to target that specific element rather than guessing a coordinate.
Where it helps most
| Situation | Why picking wins |
|---|---|
| Forms with many fields | no need to describe “the second input on the third row” |
| Dynamically rendered lists | prose is ambiguous; pointing is exact |
| Changing one specific style | the target node is unambiguous |
Next
- Browser tools — what the agent can do once you have picked.