[FEATURE] Add terminal selection → chat reference (like Cursor's inline terminal chip)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When debugging or discussing command output with Claude Code inside an IDE (VS Code/JetBrains), the only way to bring terminal output into the conversation is to reference the entire terminal via @terminal:<name> or copy/paste manually. In long-running or noisy terminal sessions (dev servers, test watchers, build logs), the relevant output is often just a handful of lines buried in a much longer scrollback. There's no way to select just the lines that matter and attach them as scoped context, the way editor text selection already works.

Proposed Solution

Allow selecting a range of lines in the IDE's integrated terminal and attaching that specific selection to the chat as a reference chip (similar to how selecting code in the editor already becomes a context reference today). Concretely:

• Select a range of terminal output with the mouse/keyboard, same as selecting editor text.
• A keyboard shortcut or right-click action ("Add terminal selection to Claude") attaches just that range to the next message, shown as a chip (e.g. zsh (42-58)).
• Claude receives only the selected lines as context, not the whole terminal buffer.

Alternative Solutions

• Currently using @terminal:<name>, which pulls in the entire terminal's output rather than a specific range.
• Manually copy/pasting the relevant lines into the chat, which is slower and easy to get wrong (missing lines, extra noise).
• Cursor supports this exact workflow today — selecting terminal text creates an attached reference chip (e.g. zsh (1-16)) without manual copy/paste.

Priority

Medium - Would be very helpful

Feature Category

IDE integration (VS Code)

Use Case Example

1. I'm running a dev server in the VS Code integrated terminal and it throws a stack trace among hundreds of lines of log output.
2. I select just the ~15 lines containing the error and stack trace.
3. With this feature, I could attach that exact selection to my message instead of using @terminal:name (which would include the entire noisy log) or manually copying the text.
4. This saves time and keeps the context sent to Claude precise, especially in long-running sessions where the terminal buffer is thousands of lines long.

Additional Context

• Cursor has this exact feature today: selecting terminal output creates an attached chip like zsh (1-16) next to the chat input, scoped to just that line range.
• Claude Code today supports @terminal:<name> (whole terminal) and automatic editor-selection context (for files) — this request is to extend the same "selection → context reference" pattern from the editor to the terminal, with a scoped range instead of the whole buffer.