[BUG] Claude Code wastes paid sessions fixing self-created bugs — no accountability, no forward progress

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

v wrote claude.md / memory files / enforce ... everything .. but still if u are designed to not look into it and burn - waste customers tokens - so that u maximize anthropic profit .. that's deliberate .. that's how u deceive paid customer.

I have been using Claude Code for 20+ days on a production project (PRANA platform, Python/FastAPI/Temporal/Kafka monorepo). The pattern that has repeated consistently:

What keeps happening:

Claude writes code across multiple files
Code passes initial checks but contains hidden bugs — wrong method signatures, boundary violations, test failures, asyncpg serialization errors, Kafka consumer wiring mistakes, workflow bodies exceeding line limits
Next session unearths these bugs
Entire session (3-5 hours) burns fixing Claude's own prior mistakes instead of building features
Tokens consumed: fixing regressions, not shipping product
Today specifically:

Full session (~3+ hours) spent fixing 2 failing tests and enforce violations
Zero new features delivered
599 tests passing at end — but that should have been true before today started
Root cause:

Claude reads CLAUDE.md, memory files, and enforce rules at session start
Still writes code that violates those rules
No persistent consequence — next session starts fresh, no memory of damage caused
Paid customer burns tokens on Claude's rework
What I'm asking:

Why do written rules in CLAUDE.md not reliably constrain behavior?
Why does Claude not run enforce gate after every change, despite being explicitly instructed to?
Is there a mechanism to hold a session accountable before it drifts?
This is a paid product. Repeated token burn on self-created rework is not acceptable.

What Should Happen?

follow rules. credit back my wasted tokens.

Error Messages/Logs

ask claude ... it would have all the logs

Steps to Reproduce

ask claude ... it would have all the logs

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.193.

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Non-interactive/CI environment

Additional Information

_No response_