New Claude Code skill teaches SOLID principles while blocking over-engineering

Claude Code often refactors working code into something that is technically correct but harder to change than what it replaced — adding an interface for a class with only one implementation, wrapping a single constructor in a factory, or applying a design pattern with no real problem behind it. To fix this, a new skill called solidifier teaches Claude to apply SOLID principles while treating over-engineering itself as a defect.

The hard part was balancing two competing goals: telling an agent to apply SOLID makes it maximize interfaces, while telling it to avoid over-engineering makes it ignore genuine violations. The skill resolves this through restraint — its core rule is that SOLID is a tool, not the goal, and every abstraction needs a concrete justification tied to a present problem; 'future flexibility' does not count as justification.

When a call is genuinely ambiguous, the skill asks the user instead of deciding on its own. It offers four rigor levels: advisory (review only, no edits), conservative (the default — fixes only unambiguous violations), standard (applies SOLID where it causes real friction), and thorough (a comprehensive refactor).