[BUG]

Environment

• Platform (select one):
• Anthropic API
• AWS Bedrock
• Google Vertex AI
• Other:
• Claude CLI version: 1.0.38 (Claude Code)
• Operating System: Windows 11 with WSL2 (Ubuntu-based Linux 5.15.153.1-microsoft-standard-WSL2)
• Terminal: WSL2 Bash

Bug Description

Claude Code has a dramatically reduced context window and is making lots of mistakes. The AI appears to have
significantly less context awareness than expected, leading to frequent errors and reduced performance in code
analysis and generation tasks.

Steps to Reproduce

1. Use Claude Code for typical development tasks (code analysis, file operations, multi-step workflows)
2. Notice reduced context retention between operations
3. Observe increased frequency of mistakes in code generation or analysis
4. Compare performance to expected baseline behaviour

Expected Behaviour

Claude Code should maintain substantial context across operations, provide accurate code analysis, and execute
multi-step tasks without significant degradation in quality or context awareness.

Actual Behaviour

Experiencing dramatically reduced context window with frequent mistakes in code operations, suggesting the AI has
less contextual information available than normal for effective development assistance.

Additional Context

• Environment: WSL2 on Windows 11 with Django 4.2 project
• Project: Complex multi-app Django application (agent-grey/Thesis Grey)
• Context: Working with large codebase with multiple interconnected components
• Impact: Reduced effectiveness for development tasks requiring context retention