[MODEL] Sonnet 4.5 false positive assumption from model for failures. Assumes failure as success, too eager to finish off

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude modified files I didn't ask it to modify

What You Asked Claude to Do

I asked it to remove a pattern from code and move code to one place. Why words to be exact "Builders are sunset, we should use the unified registry to serve instead of keeping registry in each directory"

What Claude Actually Did

Claude actually removed the example directory I gave, and said all good.

Expected Behavior

When doing something it does not reclarify, what's assuming, it just says what I think is right, but does a totally different thing, as if model is lazy

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Steps to Reproduce

_No response_

Claude Model

Sonnet

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.0.1

Platform

Anthropic API

Additional Context

_No response_