Problem
The codebase uses multiple different XML tags to wrap user content in prompts: <task>, <feedback>, <answer>, and <user_message>. This inconsistency creates unnecessary complexity in content handling and parsing logic.
Impact
- Developers: Must maintain separate detection logic for each tag type, increasing code complexity and maintenance burden
- LLM Processing: The model must handle multiple tag formats for semantically similar content, which adds cognitive overhead to prompt processing
- Code Maintainability: Changes to user content handling require updates across multiple tag detection points, increasing the risk of bugs and inconsistencies