Browser Copy

What you get

One button in a thread's header copies what you point at in a BB Browser tab. Click Copy Browser Element, pick an element in the page, and its details or picture land on your clipboard — ready to paste into a message, an issue, or an agent.

How it works

• Element text. Pick an element and get a structured markdown block: heading, the page URL and viewport, a CSS selector, ARIA role and accessible name, bounds, link, and text. This is the shape an agent reads best — it names the element and where it lives.
• Element image. Pick an element and get a PNG of just that element, clipped to its box.
• Screenshot. Copy the visible Browser tab as a PNG.

Pick a mode from the dropdown and click the element in the page. While the picker is armed the page draws a hover highlight and a label over whatever is under the cursor, so you can see the target before committing. Picking never activates the element — the click is consumed, so buttons and links do not fire.

When the thread has more than one Browser tab open, the dropdown lists them above the three modes so you can choose which tab to pick from. With one tab it goes straight to the modes.

What lands on the clipboard

Element text is written as markdown:

### a "Learn more"

**URL:** https://example.com/
**Viewport:** 1280x720
**Selector:** `html > body > p:nth-of-type(2) > a`
**Role:** link
**Name:** "Learn more"
**Bounds:** x=256, y=186, 82x18
**Link:** https://iana.org/domains/example
**Text:** "Learn more"

Only the fields that apply to the element are emitted, and free-form text keeps its full length — only the heading is trimmed for scannability.

What redaction does and does not cover

Redaction is a heuristic applied to form fields only. An <input> or <textarea> whose type is password, or whose name, id, or autocomplete attribute matches password/credential/secret/token/api-key/auth, contributes no text and no value. That covers more than passwords — a field named api_key or auth_token is caught too.

It does not cover anything that is not a form field. A token, key, or secret rendered in a <code> block, a <div>, or any other non-input element matches nothing and is copied verbatim into the markdown block, which may then go to an agent. Treat a picked element the way you would treat anything else you paste.

Images are not redacted at all: Element image and Screenshot capture whatever is on screen, so an image of a secret field is possible. Pick deliberately.

Requirements

• BB desktop with the native Browser and Plugin SDK 0.4.48 or newer. The plugin needs experimental_desktopBrowsers to attach to a tab. The header button appears only in the desktop app: picking an element means clicking it in the native Browser view, which a browser client cannot do, so the control is not offered on the web client.
• A Browser tab open in the thread you are copying from. With none open, the modes are disabled and the dropdown says so; it re-checks the thread's tabs every time you open it, so open the tab and reopen the menu.
• The clipboard is written by the page itself, so the copy needs an ordinary click with user activation — a scripted or synthetic click will be refused by the browser.

Behaviour worth knowing

• The plugin holds the tab only while you pick. It takes a short exclusive lease, refuses to interrupt automation that already controls that tab, and releases it immediately after the pick.
• Stopping control stops the pick. If you press Stop or Take over in the tab's control bar mid-pick, the pending copy ends at once, the picker stops swallowing clicks, and the button returns to its idle state.
• No side panel, no settings. The plugin contributes one header button and nothing else.
• Text and image copies are read from the page you are already looking at; nothing is uploaded, and no page content leaves the machine through this plugin.