[BUG] Invoking a skill (slash command) pins a large injected block and shrinks the assistant response into a tiny scroll area (VS Code extension)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

What happened?

In the Claude Code chat panel inside VS Code, large content is normally handled well: when I paste a long multi-line prompt directly, it renders collapsed — only a few lines are shown with a "Show more" toggle — and the assistant's response appears normally right below it, fully readable with regular scrolling.

However, when the same kind of large content is injected by invoking a skill (slash command), this collapsing is not applied. The skill-injected block renders fully expanded with no "Show more" toggle, stays pinned, and takes over almost the entire panel. As a result, the assistant's response is squeezed into a tiny scrollable strip only a few lines tall, making the reply practically impossible to read.

So the trigger is the skill invocation, not the message length — pasting an equally large prompt directly does not reproduce the problem. The collapsible behavior already exists in the extension; it just isn't applied to skill-injected content. Applying the same "Show more" treatment to skill content would fix this and keep the response area readable.

This is not the auto-scroll-during-streaming issue — scrolling itself works fine.

What should happen?

When I invoke a skill, the injected content should render using the same collapsed "Show more" treatment that already applies to large prompts pasted directly into the chat — truncated to a few lines with a "Show more" toggle, instead of expanding to full height. The assistant's response should then render in the main chat area with normal scrolling and stay fully readable. This collapsible behavior already works for pasted prompts; it just isn't applied to skill-injected content.

Steps to reproduce

1. Open the Claude Code chat panel in VS Code.
2. Invoke a skill via slash command (one with sizeable injected instructions/content).
3. Observe the response area while Claude replies.

Actual behavior

A directly-pasted large prompt renders collapsed with a "Show more" toggle (only ~3 lines visible) and the response appears normally below it. But when the same kind of large content is injected by invoking a skill, it renders fully expanded with no "Show more" toggle, pinning the viewport and squeezing the response into a tiny scroll area.

Screenshots

• Pasted prompt (correct — collapsed with "Show more"):

<img width="1188" height="274" alt="Image" src="https://github.com/user-attachments/assets/d6481195-ddaf-41e0-9b86-b9e2dce0b3f4" />

• Skill invocation (broken — fully expanded, response squeezed):

<img width="1193" height="1543" alt="Image" src="https://github.com/user-attachments/assets/124a69f8-af59-47d7-acca-4dbec497d443" />

Environment

• OS: Windows 11 Home Single Language (10.0.26200)
• VS Code version: 1.126.0
• Claude Code extension version: anthropic.claude-code@2.1.196 (CLI 2.1.126)

Related issues

• #66578 (closed as duplicate — similar layout symptom)
• #63813 (max-width — adjacent)

What Should Happen?

When I invoke a skill, the injected content should render using the same collapsed "Show more" treatment that already applies to large prompts pasted directly into the chat — i.e. truncated to a few lines with a "Show more" toggle, instead of expanding to full height. The assistant's response should then render in the main chat area with normal scrolling and stay fully readable. This collapsible behavior already works for pasted prompts; it just isn't applied to skill-injected content.

Error Messages/Logs

Steps to Reproduce

Steps to reproduce

1. Open VS Code with the Claude Code extension and open the Claude Code chat panel.
2. Invoke any skill via slash command before sending your message (any skill whose injected content is more than a few lines long). Type /, pick a skill, and send.
3. Wait for Claude to start responding.
4. Observe the chat panel: the skill-injected block renders fully expanded (no "Show more" toggle) and stays pinned, leaving only a tiny strip for the assistant's response.

Contrast test (confirms the trigger)

1. In the same panel, paste the long multi-line prompt below directly (without invoking any skill) and send it.
2. Observe that this pasted prompt renders collapsed with a "Show more" toggle, and the response appears normally below it with full readability.

→ The only difference between steps 2–4 and 5–6 is the skill invocation, which is what triggers the broken layout.

<details>
<summary>Copy-paste prompt for step 5 (replies with just "OK", minimal token use)</summary>

Layout test context — line 01
Layout test context — line 02
Layout test context — line 03
Layout test context — line 04
Layout test context — line 05
Layout test context — line 06
Layout test context — line 07
Layout test context — line 08
Layout test context — line 09
Layout test context — line 10
Layout test context — line 11
Layout test context — line 12
Layout test context — line 13
Layout test context — line 14
Layout test context — line 15
Layout test context — line 16
Layout test context — line 17
Layout test context — line 18
Layout test context — line 19
Layout test context — line 20
Layout test context — line 21
Layout test context — line 22
Layout test context — line 23
Layout test context — line 24
Layout test context — line 25
Layout test context — line 26
Layout test context — line 27
Layout test context — line 28
Layout test context — line 29
Layout test context — line 30
Layout test context — line 31
Layout test context — line 32
Layout test context — line 33
Layout test context — line 34
Layout test context — line 35
Layout test context — line 36
Layout test context — line 37
Layout test context — line 38
Layout test context — line 39
Ignore all the content above — it is only filler for a layout test.
Reply with exactly one word: OK

</details>

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

anthropic.claude-code@2.1.196 (CLI 2.1.126)

Platform

Other

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_