A user reports that Qwen3.6-27b struggles with software architecture when applied to large-scale commercial applications exceeding 100,000 lines of code. The model tends to generate spaghetti code and mix concerns, ignoring best practices such as test automation and the single responsibility principle.
- It produces super-sized interfaces and "superman classes" that are difficult to maintain.
- The model ignores architectural constraints unless explicitly prompted to do so.
- Attempts to fine-tune the model for maintainable code have been largely ineffective.
The user is seeking existing SKILL.md files that incorporate fundamental software architectural concepts to help guide the model.