[FEATURE] critical thinking mode

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 is fantastic and offers "thinking modes" via "think", "think harder", or "ultrathink". But these are thinking levels. So it the new thinking effort thing. What is missing for me are directives and modes about "how it thinks" and how it interacts with the user.

Proposed Solution

In my CLAUDE.md I require Claude to think critically and to avoid complimenting me all the time. In fact I tell it that whenever it is about to compliment me, to think critically and challenge me instead. In short, I want Claude to enhance my own thinking and to be a better "space to think" as a result. I like the overall disposition as an agent, but it is a bit to eager to please sometimes, and takes prompts sometimes without sufficient criticality. I want it to always approach prompts as opinions, not only as instructions. Like a wise coworker or subordinate who keeps challenging your o thinking in a constructive and respectful way. If it makes sense to it it would proceed without further comment, but if it thinks there's a better way ot that the insuruction makes not sense, or can have bad side-effects then it should say so.

Another think that I keep having to repeat is phrases like "ask me any clarifying questions before implementing. if you have none then proceed". This would be also a critical thinking behavior in my book. Don't implement unless everything is crystal clear.

Another problem with putting all this is CLAUDE.md is that the material in CLAUDE.md does not seem to be managed in the context any differently than any other interaction. So it dilutes over time, even without compaction, and sooner or later, Claude basically forgets that you asked it to think critically. So i constantly find myself having to be explicit about it. I even created a slash command to remind it every once in a while.

So What I would want is a "critical thinking mode" that would be independent of the thinking effort level - it's about a thinking attitude. You could even have levels for this mode - but just having it would be a great enhancement to its behavior. When I ask it to think critically it does much better than when I don't.

Alternative Solutions

An alternative, or a completely separate feature, would be to make CLAUDE.md more "special" so that it does not get diluted (with or without compaction). In the context management give it some kind of prominence so that it is followed by Claude more closely. It seems to me that this would be a more complex solution and not one that focuses specifically on "how it thinks"

Priority

High - Significant impact on productivity

Feature Category

Configuration and settings

Use Case Example

I'm working on a new feature or enhancement. I have an idea what needs to be done. I provide a prompt or a specification. Claude considers it but digs a little deeper. If i didn't provide the motivation for the feature it might ask me why we need this. if it thinks that there may be a better approach it would tell me about it and ask whether I want to work together on exploring that. It would be particularly sensitive to undesired side-effects and security issues.

When I give it vague or not fully defined requirements it would automatically ask for clarifications.

Additional Context

_No response_