Feature request: allow Copilot agents to read user-opened terminal panes with...

Start at the /chat terminal-pane experience and review how terminal actions such as “Send terminal to chat” or Ctrl+L could be represented in the app. Done means explicit, visible, time-limited, terminal-scoped and revocable consent lets Copilot read the approved terminal context without default access.

Feature request: allow Copilot agents to read user-opened terminal panes with explicit consent.

Use case:
I opened a terminal manually inside the GitHub Copilot app and ran an external agent CLI (ant). When I asked Copilot if it could see the terminal output, it could detect that a terminal panel existed from context, but it could not read the scrollback because the terminal instance was not accessible to the agent.

Desired behavior:

• User opens a terminal manually and runs commands.
• User asks Copilot to inspect the terminal, or clicks a terminal action such as “Send terminal to chat”.
• App asks for explicit per-terminal consent, or the click itself acts as an intentional share action.
• If approved, Copilot can read current viewport / recent scrollback / selected scrollback / full scrollback.
• Access should be time-limited, visible, scoped to that terminal, and revocable.

Prior art / UX reference:
In Antigravity IDE, when a terminal is open, the top bar has an @ action, also available via Ctrl+L, labeled “Send terminal to chat”. This is a useful interaction pattern: the user explicitly chooses to attach/share the terminal context with the chat instead of the agent having automatic access.

Why this matters:
This would improve debugging and coordination when using external agent CLIs (ant, agy, copilot) inside Copilot app project worktrees. It would avoid manual copy/paste of terminal output while preserving the right privacy model: no terminal access by default, only explicit user-initiated sharing.