Auto Mode: should only auto-approve tool calls, not inject behavioral overrides

Problem

When I enable Auto Mode, a system prompt is silently injected that changes the AI's reasoning behavior:

1. Execute immediately — Start implementing right away. Make reasonable assumptions and proceed on low-risk work.
2. Minimize interruptions — Prefer making reasonable assumptions over asking questions for routine decisions.
3. Prefer action over planning — Do not enter plan mode unless the user explicitly asks.
4. Expect course corrections
5. Do not take overly destructive actions
6. Avoid data exfiltration

My intent when enabling Auto Mode is simply to skip the tool permission/approval prompts — not to fundamentally change how the AI reasons, plans, or communicates.

Observed issues

These injected instructions cause the AI to:

• Rush into implementation without confirming ambiguous requirements
• Skip planning steps that would otherwise be appropriate
• Make assumptions instead of asking clarifying questions
• Acknowledge mid-response that it should confirm with the user, but never actually pause — it immediately proceeds because "Auto Mode is active"
• Frequently take unsolicited next steps beyond what was asked

Suggestion

Auto Mode should only affect the tool permission layer (auto-approve tool calls), without injecting any system-level behavioral instructions. Let my CLAUDE.md and conversation context control the AI's working style. If behavioral hints must be kept, they should be opt-out or separately configurable from the permission setting.