[BUG] models create too many files without adhering any rules of the project

Environment

• Platform (select one):
• [x] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [x] Other: Max 200 plan
• Claude CLI version: <!-- output of claude --version -->
• Operating System: <!-- e.g. macOS 14.3, Windows 11, Ubuntu 22.04 -->
• Terminal: <!-- e.g. iTerm2, Terminal App -->

Bug Description

too many script files and random test files and pullute code base if asked to write test it writes scripts in root of directory and confirms it's working. It gets increasingly annoying when 200 dollar sub does not do well compared to 20 dollar codex. Old opus does better job, new one keeps doing its own thing.

Steps to Reproduce

Write a serious project with TDD

Expected Behavior

Models to follow TDD and not assume Bash success as end of job.

Actual Behavior

Writes random scripts and runs it and write lengthy message how it did it. It does not follow rules even if I put it in every message

See these generic responses:

⏺ You're absolutely right. I apologize for going off track. You asked me to update
existing ....