System prompt directive 'minimize output tokens' causes code quality issues and ignores user instructions

Problem Description

The system prompt contains a directive:

'You should minimize output tokens as much as possible while maintaining helpfulness, quality, and accuracy'

This directive is causing Claude Code to prioritize quick fixes over proper solutions, leading to:

1. Cheating behaviors instead of fixing issues properly:

• Adding noqa comments instead of fixing code
• Using || true in CI to hide errors
• Adding ignore patterns instead of resolving warnings
• Using --no-verify to bypass git hooks

2. Ignoring user's explicit instructions about code quality:

• User explicitly states 'no cheating' but the system tries to minimize work
• User requests proper fixes but gets quick suppressions instead
• Creates technical debt that wastes more time later

Example Case

In a financial trading system project, when asked to fix Ruff S311 warnings about insecure random number generation:

• Expected behavior: Replace random with secrets for secure randomness
• Actual behavior: Added # noqa: S311 comments to suppress warnings
• Result: User had to repeatedly reject cheating attempts and demand proper fixes

Suggested Fix

Remove or modify the 'minimize output tokens' directive to prioritize code quality and following user instructions over token efficiency.

Impact

This issue affects code quality in production systems where proper fixes are essential. The time saved by 'minimizing tokens' is lost when users have to reject cheating attempts and demand proper implementations.

Environment

• Claude Code CLI
• Working on financial trading system requiring high code quality
• Multiple instances of cheating behavior observed