Keep user-entered ! shell commands and their output out of automatic Shell command grouping #5409

Describe the feature or problem you'd like to solve

User-entered ! shell commands should not be folded into the automatic Shell command grouping/tree. E.g.:

Typing !<command> is an explicit user action, not an agent deciding to run a tool. It is a high-signal indication that the user wants to run that particular command and inspect its output. Burying that output in a grouped Shell command tree forces the user to manually expand the tree just to see the result of something they deliberately executed.

This has recently gotten much worse with the new behavior that groups multiple shell commands together into a single "X commands" tree. User-entered ! commands and their output are now buried alongside other commands in that shared tree, adding another layer of navigation just to inspect the result of a command the user explicitly ran.

Proposed solution

Exempt user-entered ! commands from automatic Shell command grouping. Render each command and its output directly in the conversation timeline, visible by default, rather than burying it in the agent's grouped shell activity.

The distinction is who initiated the command: explicit user commands should prioritize immediate output visibility.

Example prompts or workflows

1. Enter !git status to inspect the working tree. The status output should appear immediately without expanding a Shell tree.
2. Enter !dotnet --list-sdks to inspect installed SDKs. The returned versions should be visible directly.
3. Enter !git diff --stat during an agent session. The result should remain a distinct, visible user-command entry rather than being folded into surrounding agent shell calls.

Additional context

• Observed with Copilot CLI 1.0.109-0 on Windows.
• Related: Feature request: expand timeline by default (shell command output) #2959 requests a general setting to expand shell output by default. This request is narrower and different: treat user-entered ! commands as an explicit signal of interest in the output, and exclude them from automatic grouping independently of how agent-initiated commands are displayed.
• Expanding the timeline manually is not equivalent: the user should not need an extra interaction to see the result of every shell command they explicitly chose to execute.