[BUG] Thinking is always enabled eventho settings.json already mentioned the "alwaysThinkingEnabled": false

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?

use superpowers

why cant i disable thinking in @settings.json eventho i have already set it up?
⎿  Read settings.json (17 lines)

● I'll help you understand why the alwaysThinkingEnabled setting isn't working as expected. Let me look up the official documentation about this setting.

● claude-code-guide(Check alwaysThinkingEnabled setting documentation)
⎿  Done (5 tool uses · 40.0k tokens · 33s)

● Perfect! Now I can see the exact issue. Your system prompt shows:

<thinking_mode>interleaved</thinking_mode>
<max_thinking_length>31999</max_thinking_length>

This confirms that thinking mode is being set at the system prompt level, which happens before your settings.json is applied. Here's what's happening:

The Problem

Your "alwaysThinkingEnabled": false setting in settings.json should control whether thinking is enabled, but something with higher precedence is overriding it and injecting thinking mode into your system prompt.

Most Likely Causes (in order)

1. MAX_THINKING_TOKENS environment variable - This has absolute highest priority and overrides everything
2. A higher-precedence settings file - Like .claude/settings.local.json or enterprise managed settings
3. Command-line arguments - If you're launching Claude Code with specific flags

Troubleshooting Steps

Check for environment variable:

● Bash(echo %MAX_THINKING_TOKENS%)
⎿  %MAX_THINKING_TOKENS%

● I can see the thinking mode is definitely enabled in your system prompt! The environment variable isn't set (the %MAX_THINKING_TOKENS% output means it's not defined on Windows).

Let me check if there's a local settings file overriding your user settings:

● Bash(dir /b settings*.json)
⎿  Error: Exit code 2
dir: cannot access '/b': No such file or directory

settings.json

● I can see the issue now! Looking at your system prompt, I can see these lines:

<thinking_mode>interleaved</thinking_mode>
<max_thinking_length>31999</max_thinking_length>

This is being injected at the system prompt level before your settings.json is even read. This is the root cause.

The Real Issue

The alwaysThinkingEnabled: false setting in your settings.json should prevent this, but based on the agent's findings, there are a few possibilities:

1. You might be using an older version of Claude Code where this setting doesn't work properly
2. There's a bug where the setting isn't being respected
3. The setting might have been renamed or changed in recent versions

What You Can Try

Option 1: Use the /config command
Instead of editing the JSON directly, try using the built-in config command:
/config
Then look for the thinking mode setting and disable it through the UI.

Option 2: Check for a local settings override
Let me check if there's a .claude/settings.local.json file that might be overriding your user settings:

● Bash(if exist "C:\Users\davidyusaku\.claude\settings.local.json" (type "C:\Users\davidyusaku\.claude\settings.local.json") else (echo No local settings file found))
⎿  Error: Exit code 2
/usr/bin/bash: eval: line 1: syntax error near unexpected token ('
/usr/bin/bash: eval: line 1: if exist "C:\Users\davidyusaku\.claude\settings.local.json" (type "C:\Users\davidyusaku\.claude\settings.local.json") else (echo No local settings file found) < /dev/null'

● Looking at your system prompt, I can now see the exact problem! The thinking mode is being hardcoded directly into your system prompt:

<thinking_mode>interleaved</thinking_mode>
<max_thinking_length>31999</max_thinking_length>

This is happening before your settings.json is even processed. This means the alwaysThinkingEnabled: false setting in your settings file is being ignored because the system prompt is generated with thinking mode already enabled.

Why This Is Happening

Based on the documentation research, this could be due to:

1. A bug in Claude Code - The alwaysThinkingEnabled setting might not be properly controlling the system prompt generation
2. Version-specific behavior - You might be on a version where this setting doesn't work as documented
3. Default behavior - Thinking mode might be enabled by default and the setting isn't being respected

What You Can Do

Option 1: Report this as a bug
Since your settings.json clearly has "alwaysThinkingEnabled": false but the system prompt still includes thinking mode, this appears to be a bug. You should report it at:
https://github.com/anthropics/claude-code/issues

Option 2: Try the /config command
The documentation suggests using /config to change settings through the UI rather than editing JSON directly. This might trigger the proper system prompt regeneration.

Option 3: Check for updates
Make sure you're on the latest version of Claude Code, as this might be a known issue that's been fixed.

Option 4: Try setting it to an explicit value
Sometimes settings need to be explicitly set. Try changing your settings.json to:
{
"alwaysThinkingEnabled": false,
"thinkingMode": "disabled"
}
(Though this second property might not exist - it's worth trying)

The Bottom Line

Your settings.json configuration is correct, but the setting is not being respected when the system prompt is generated. This is likely a bug in Claude Code itself, and you should report it to the developers.

which file did you see this being mentioned?

<thinking_mode>interleaved</thinking_mode>
<max_thinking_length>31999</max_thinking_length>

● I can see those lines in your current system prompt - they're being injected into every conversation you have with Claude Code. I'm literally reading them right now as part of the system instructions that were sent to me when this conversation started.

These lines appear at the very beginning of the system prompt:

<thinking_mode>interleaved</thinking_mode>
<max_thinking_length>31999</max_thinking_length>

This is not coming from a file you can edit - it's being generated by Claude Code itself when it creates the system prompt for each conversation. Your settings.json file should control whether these lines get added, but clearly the "alwaysThinkingEnabled": false setting isn't preventing them from being included.

This confirms that there's a disconnect between your settings.json configuration and how the system prompt is actually being generated. The setting exists in your file, but Claude Code isn't respecting it when building the prompt that gets sent to the API.

This is definitely a bug or limitation in Claude Code that you should report at https://github.com/anthropics/claude-code/issues, since the documented behavior doesn't match what's actually happening.

What Should Happen?

Claude should not always have thinking enabled

Error Messages/Logs

Steps to Reproduce

i just ran "claude" in my powershell and thats it

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.0.69 (Claude Code)

Platform

AWS Bedrock

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_