[FEATURE] Decouple terminal output verbosity from context persistence (Separate human-facing output from model working memory).

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

Claude Code currently uses the same assistant messages for two different purposes:

Communicating with the user.
Preserving working memory for future turns.

As a result, every explanatory message ("I've inspected...", "I noticed...", "Next I'll...") becomes part of the conversation history and contributes to context growth.

For long-running development sessions this causes the context window to fill with natural-language explanations whose primary purpose was human readability rather than future reasoning.

This is especially noticeable during repository exploration, code reviews, and debugging sessions where Claude performs many tool calls and narrates each step.

Proposed Solution

Allow Claude Code to separate:

• Presentation output (shown to the user)
• Persistent conversational context (fed back to the model)

For example, I would like to keep the terminal output verbose and educational while storing only a compact working summary in future context.

<img width="830" height="401" alt="Image" src="https://github.com/user-attachments/assets/8fddafdf-8a71-445b-9a36-9ddb98e17978" />

<img width="810" height="404" alt="Image" src="https://github.com/user-attachments/assets/17282dba-ec89-4daf-a055-6f0fe0fa9609" />

<img width="802" height="444" alt="Image" src="https://github.com/user-attachments/assets/735f4565-3442-48e7-88ad-0a7fd0c6b1eb" />

<img width="806" height="498" alt="Image" src="https://github.com/user-attachments/assets/33d9f9a6-7dbf-416c-8849-28fb0028ae0e" />

Alternative Solutions

I've tried to use /compact command with custom instructions, but it quie often just loses important details.

Priority

High - Significant impact on productivity

Feature Category

Performance and speed

Use Case Example

<img width="801" height="690" alt="Image" src="https://github.com/user-attachments/assets/b4dde941-7957-481a-9e5d-d73639aa8fe9" />

Additional Context

I primarily use Claude Code for large software projects (Salesforce/Apex/LWC, Azure DevOps, large pull requests), where sessions often involve hundreds of tool invocations and many hours of interaction. In these workflows, explanatory assistant messages can become a significant portion of the accumulated conversation history despite providing little long-term value compared to the conclusions themselves.