Feature Request: Display Model Reasoning and Adjust Reasoning Intensity

[Johnson1662]

Feature Description

As a terminal command-line tool, I would like to suggest adding the following two features:

1. Display Model Reasoning: When generating code or explaining commands, the tool should be able to show the model's thinking process, reasoning chain, or intermediate steps, helping users understand how the model arrived at its conclusions.

2. Adjust Reasoning Intensity: Allow users to adjust the model's reasoning intensity/depth via parameters (e.g., --reasoning-level or -r), such as:

   • low: Quick response with fewer reasoning steps
   • medium: Balanced speed and quality
   • high: Deep reasoning with full thinking process displayed

Motivation

• Transparency: Users can see how the model analyzes problems and generates code
• Debugging Aid: When output doesn't meet expectations, the reasoning process helps identify issues
• Flexible Control: Different scenarios require different trade-offs between response speed and quality

Proposed Solution

# Display reasoning process
command-code "implement quicksort" --show-reasoning

# Adjust reasoning intensity
command-code "optimize this code's performance" --reasoning-level high

# Combined usage
command-code "analyze this bug" -r high --show-reasoning

Alternatives Considered

• Use environment variable control (e.g., COMMAND_CODE_REASONING_LEVEL)
• Set default reasoning intensity in configuration file

Looking forward to your feedback!

[ahmadbilaldev]

Thanks for the feedback. We are shipping configurable reasoning effort very soon. As far as reasoning/thinking goes, we already show it for models that support it.

I am curious to know if you intentionally want these in the non-interactive headless mode, or is this just an example?

[Johnson1662]

It is just an example.Most of the time, the model I use is DeepSeek V4 Pro. As we know, this model's capabilities vary significantly depending on the intensity of its reasoning; this is precisely why I am interested in the feature that allows for adjusting inference intensity.

[vipulgupta2048]

Just added support --model for #383