[FEATURE] Add ability to control tone/language style used in “extended thinking” traces (internal reasoning text)

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

Summary

Claude Code currently allows users to control the tone and style of visible responses through CLAUDE.md (global or project‑level). However, extended thinking traces—the internal chain‑of‑thought that Claude displays in the IDE—do not follow these instructions.

This results in internal thinking that sometimes uses whimsical, metaphorical, or overly dramatic language (e.g., “forging ahead,” “frolicking through ideas,” etc.), which can be distracting and inconsistent with the desired professional tone.

There is currently no configuration (project-level, global, CLI, or IDE) that allows users to control or constrain extended thinking style.

***

Problem / Impact

• Breaks immersion in professional coding environments.
• Disrupts focus during complex debugging or planning tasks.
• Causes tone mismatches when visible responses are strictly controlled via CLAUDE.md.
• Makes extended reasoning appear less reliable or less serious to some users.
• Creates a cognitive disconnect between “thinking” and “output,” even though both are part of the user experience.

The inability to control extended thinking also prevents teams from enforcing consistent communication standards.

***

What I Expected

That extended thinking would:

1. Respect at least the same tone rules defined in CLAUDE.md
2. Or expose a separate setting to control reasoning‑trace style
3. Or allow extended thinking to be fully disabled, collapsed, or simplified (e.g., “short technical reasoning only”)

***

Actual Behavior

Extended thinking ignores tone/style guidance entirely and continues using:

• Metaphors
• Poetic verbs
• Dramatic narrative tone
• Whimsical or playful language

Even when the visible output is strictly technical and plain.

***

Why This Matters

For many developers, Claude Code is a tool used in:

• Professional environments
• Enterprise codebases
• Security-sensitive workflows
• Highly technical collaborations

Having playful or poetic reasoning output makes the tool feel less precise, even when the code diffs are excellent.

Tone control here would help:

• Improve clarity
• Reduce distraction
• Build trust
• Match organizational standards
• Align reasoning with visible output tone

Proposed Solution

Proposed Solutions

Any of the following would solve the issue:

Option A — Follow CLAUDE.md automatically

Respect tone/style rules for internal reasoning just as visible output does.

Option B — New extendedThinkingStyle setting

Example:

{
  "extendedThinkingStyle": "technical | concise | neutral | minimal | default"
}

Option C — Add a toggle

• Show Simplified Thinking
• Hide Extended Thinking by default
• Full Extended Thinking
• Technical‑only Extended Thinking

Option D — Project-level override

Allow per‑project tone control (similar to CLAUDE.md) specifically for reasoning traces.

***

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

Reproduction Steps

1. Create strict tone rules in ~/.claude/CLAUDE.md (e.g., “no metaphors, no whimsical language”).
2. Open Claude panel in VS Code.
3. Ask Claude to perform a multi-step reasoning-heavy task.
4. Observe:

• Visible output follows rules
• Extended thinking does not

Additional Context

Request

Please add tone/style configuration for extended thinking, or allow users to disable or simplify it. This would make Claude Code significantly more consistent, professional, and aligned with user expectations.