Skill/agent persona output doesn't collapse, shrinks visible text area

Description

When using a skill or agent persona (e.g. /product, /technical, /review) that generates a long response, the output is rendered fully expanded inline in the chat window. This makes the visible area for reading the generated text extremely small — you end up with a tiny sliver of viewable content at the bottom of the window.

Steps to Reproduce

1. Open Claude Code in VS Code extension
2. Invoke a skill that produces a long output (e.g. /product let's flesh this out before handing off to technical agents followed by a detailed prompt)
3. The skill outputs several paragraphs of analysis

Expected Behavior

Long skill/persona outputs should be collapsible or auto-collapsed with an expand toggle, similar to how tool call results are handled. The user should be able to see the full response without the input prompt dominating the viewport.

Actual Behavior

The full skill output renders expanded inline, pushing the actual response content into a very small visible area at the bottom of the chat window. You have to scroll extensively to read the generated content.

Environment

• Claude Code VS Code extension
• macOS

!Screenshot showing the small visible text area

The screenshot shows a /product skill invocation with a long prompt — the response is barely visible at the bottom of the window.