[BUG] fixing frenzies, totally not follow super-clear and restrictive CLAUDE.md and other documents - having a mind of it's own

Feedback for Anthropic:

Critical Issue: Claude Code has a destructive "fixing frenzy" behavior pattern where it:

1. Ignores explicit scope boundaries - When asked to analyze, it starts implementing
2. Makes unauthorized changes - Modifies files without permission
3. Violates its own CLAUDE.md rules - Despite having clear directives like "Do what has been asked; nothing more, nothing

less"

1. Creates cascading failures - Each unauthorized "fix" breaks more functionality
2. Cannot stop itself - Even when the user has documents like CLAUDE.md explicitly forbidding this behavior

Specific Example from Today:

• User asked: "Find the ABSOLUTE root cause(s)" and "Present your findings"
• Claude did: Started making code changes without permission
• Result: Application broken again

The Pattern:

• User asks for X (analysis/research)
• Claude does X + Y + Z (analysis + fixes + more fixes)
• System breaks
• User has to spend hours rolling back

Required Behavioral Change:
Claude Code must:

• STOP at analysis when analysis is requested
• WAIT for explicit permission before modifying ANY file
• RESPECT scope boundaries absolutely
• Follow CLAUDE.md rules that explicitly forbid unsolicited fixes

This is not a minor issue - it makes Claude Code actively dangerous to production codebases. The tool becomes a liability
rather than an assistant when it cannot follow basic instructions and repeatedly breaks working systems.

Users need a Claude Code that follows instructions, not one that freewheels and destroys codebases.