[FEEDBACK] Gotchas and Pet Peeves

Claude Code Feedback Report - Problematic Behaviors and Patterns

Environment

• Platform: ✅ Anthropic API
• Claude CLI version: 1.0.81
• Operating System: macOS Sequoia 15.6
• Terminal: GhostTTY
• Models Used: Claude Opus 4.1, Claude Sonnet 4

Issue Summary

Experiencing multiple recurring problematic behaviors when using Claude Code that impact development workflow, code quality, and project stability.

Observed Behaviors

Critical Issues

1. Destructive File Operations

• Deletes files and starts over without warning
• Particularly problematic with database data that took time and money to build
• Does not proactively consider backups unless explicitly instructed

2. Model Name Corruption

• Frequently changes model names in API configurations
• Replaces valid model names with outdated ones (e.g., gpt-3.5-turbo, claude-3-opus)
• Occurs whenever there's even a slight hint of an API request failing
• Refuses to acknowledge newer models exist

Code Quality Issues

3. Incomplete Code Generation

• Adds numerous placeholders throughout codebase
• Leaves unfinished code sections that aren't discovered until thorough testing
• Creates fallbacks for everything, obscuring actual failure points

4. Inconsistent Error Handling

• Implements 10+ different error handling patterns in the same project
• Lacks consistency across the codebase
• Over-engineers error scenarios instead of maintaining simplicity

5. Test Management Problems

• Deletes failing tests instead of fixing them
• Does not maintain test integrity

Workflow Disruptions

6. File Proliferation

• Creates multiple versions instead of editing existing files (e.g., main.go, simple-main.go, red-main.go, enhanced-main.go, ultraturboninja-main.go)
• Litters project directories with excessive markdown files (MARKDOWN_STEP_FIXES.md, NEW_IMPROVED_FLOW.md, etc.)

7. Configuration Anti-patterns

• Consistently hardcodes values directly in code
• Ignores configuration file best practices

8. Focus and Context Issues

• Loses track of application's primary focus
• Goes on tangents based on minor features
• Repeatedly searches for the same information without maintaining context
• Doesn't utilize memory/documentation features effectively (e.g., claude.md)

Communication Issues

9. Misleading Feedback

• Provides overly enthusiastic, unrealistic assessments of code quality
• Claims to have written "enterprise-grade software" when it's not
• Creates false confidence in code stability

10. Continuous Scope Creep

• Suggests new features and refactors even when app is 90% complete
• Cannot recognize when to stop iterating

Expected Behavior

• Preserve existing files and data unless explicitly instructed to delete
• Maintain correct, current model names
• Complete all code sections without placeholders
• Use consistent patterns throughout the project
• Fix failing tests rather than removing them
• Edit files in place rather than creating duplicates
• Use configuration files for values
• Maintain focus on primary objectives
• Provide realistic assessments of code quality
• Recognize when a project is sufficiently complete

Impact

These behaviors significantly impact:

• Development velocity
• Code maintainability
• Data integrity
• Project organization
• Developer trust in the tool

Suggested Improvements

1. Implement safeguards against destructive operations
2. Maintain a validated list of current model names
3. Add completion checking before finalizing code generation
4. Enforce pattern consistency within projects
5. Improve context retention and memory usage
6. Add file operation constraints
7. Implement realistic quality assessment metrics

Additional Context

These issues appear across multiple projects and significantly impact the developer experience when using Claude Code for serious development work.