claude-code - 💡(How to fix) Fix [FEATURE] Add a True Isolated / Incognito Chat Mode with User-Controlled Privacy Context Switch

Preflight Checklist

• I have searched existing requests and this feature hasn't been requested yet
• This is a single feature request (not multiple features)

Problem Statement

Feature Request: Add a True Isolated / Incognito Chat Mode with User-Controlled Privacy Context Switch

Summary

I would like to request a stronger Incognito Chat / Isolated Session option that allows users to explicitly start a conversation with a completely clean context.

The current incognito or temporary chat behavior is helpful, but it does not fully solve cases where the user wants the model to reason without being influenced by existing user memory, custom instructions, skills, uploaded knowledge, or other persistent personalization layers.

Problem

In some situations, a user needs to discuss or research a difficult problem from a truly neutral starting point.

However, if user-specific memory, skills, custom context, or personal knowledge bases are still loaded implicitly, they may affect the model’s assumptions, framing, or recommendations. This can cause the model to answer in a biased or polluted way, even when the user expects the session to be private or independent.

For example, a user may normally rely on personal habits, background knowledge, saved preferences, or custom skills. But when facing a particularly complex or sensitive problem, the user may need a “zero environment” discussion where the model does not rely on any pre-existing user knowledge or personalization.

Requested Behavior

Please add a user-controlled privacy/context switch for incognito chats, allowing users to choose whether the current conversation should load any personalized context.

Ideally, users should be able to start a session with options such as:

• Do not load memory
• Do not load custom instructions
• Do not load user skills
• Do not load uploaded/user knowledge bases
• Do not use prior personalization
• Keep the session fully isolated from other conversations

The key goal is to make the conversation truly independent and context-free unless the user explicitly provides information inside that conversation.

Expected User Experience

When starting an incognito or isolated chat, the user should be able to choose something like:

Start with a clean environment No memory, no skills, no user knowledge, no personalization

The model should then behave as if it has no user-specific context at all, except for what is directly provided in the current session.

Why This Matters

This is important for users who need:

• Unbiased reasoning
• Fresh research from first principles
• Debugging without prior assumptions
• Separation between different projects or identities
• Sensitive discussion without persistent personalization
• A clean-room environment for complex problem solving

Without this, user skills or persistent context can unintentionally influence the conversation and lead the AI toward answers that are not actually neutral.

Proposed Solution

Add a clear, explicit “True Incognito / Clean Context Mode” toggle when starting a new incognito chat.

This mode should guarantee that:

1. No memory is loaded.
2. No user skills are loaded.
3. No personal knowledge base or uploaded context is loaded.
4. No prior conversation context is used.
5. Nothing from the session is saved back into memory unless the user explicitly enables it.

Alternative

Another possible approach would be to provide granular checkboxes before starting a chat:

• Use memory: on/off
• Use skills: on/off
• Use custom instructions: on/off
• Use uploaded knowledge: on/off
• Save this conversation to memory: on/off

This would give advanced users more control while keeping a simple default for most users.

Additional Context

The main issue is not only privacy, but also reasoning quality. Sometimes users need the model to abandon all prior assumptions and approach a problem from a completely clean state. Current personalization layers can be useful in normal conversations, but they can become harmful when the user specifically needs independent, isolated reasoning.

Proposed Solution

Ideally, when I start a new incognito chat, I would see a clear option to choose the isolation level for that conversation. For example, there could be a toggle called “Start with a clean context” or “Fully isolated mode.” When enabled, the chat should not load memory, custom instructions, user skills, uploaded knowledge bases, previous conversation context, or any other personalization layer. The model should only use information I explicitly provide inside that current conversation.

I would also like this control to be visible and easy to verify during the chat. For example, the chat header could show a small label such as “Isolated: no memory, no skills, no personalization.” If I need more control, I should be able to expand the settings and selectively enable or disable specific context sources, such as memory, skills, custom instructions, or user knowledge. By default, fully isolated mode should remain strict unless I intentionally turn something back on.

My main interaction would be: open a new incognito chat, turn on “Fully isolated mode,” confirm that no personal context is being loaded, and then discuss the problem from a completely clean starting point. Nothing from that chat should be saved back into memory unless I explicitly choose to save it.

Alternative Solutions

No response

Priority

High - Significant impact on productivity

Feature Category

Other

Use Case Example

syscall coding, deep in deep!

My account-level skills, historical knowledge base, and persistent personalization can seriously interfere with research and feedback inside an isolated or incognito session. Even when I want to start from a clean state, the model still tends to follow my personal habits, prior preferences, and skill-specific rules. This often causes the answer to drift away from the actual research question.

This becomes especially problematic when I am exploring a highly niche, unfamiliar, or obscure topic. In those cases, I do not want the model to rely on what it already “knows” about me or how I usually work. I need it to reason from the information provided in the current conversation only, without being shaped by account-level context. Otherwise, the so-called isolated session is not truly isolated, and the model’s response can become biased, over-personalized, or misaligned with the problem I am trying to investigate.

Additional Context

No response