[MODEL] Utter Shit instruction following and learning from past patterns.

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

Session ID 235ff9c5-2a96-48e1-87f6-e7617e4e15f2,

What Claude Actually Did

Just did stupid things, made wrong assumptions, bullshitted about RCAs after multiple prompts of saying "look at upstream/downstream impact to a full trace", made wrong choices, latches on to the first ever assumption that it can make, just a piece of shit.

Expected Behavior

Session ID 235ff9c5-2a96-48e1-87f6-e7617e4e15f2,

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

Session ID 235ff9c5-2a96-48e1-87f6-e7617e4e15f2,

Claude Model

Opus

Relevant Conversation

Session ID 235ff9c5-2a96-48e1-87f6-e7617e4e15f2,

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.1.90

Platform

Anthropic API

Additional Context

Session ID 235ff9c5-2a96-48e1-87f6-e7617e4e15f2,