openhands/score-style

Evaluate migrated code against style guidelines and best practices.

This skill scores code quality aspects that are independent of functional correctness — focusing on readability, maintainability, and adherence to conventions.

Default Scoring Criteria

Unless a custom rubric is provided, score on these attributes (1-5 scale):

Naming Conventions (1-5)

• 1: Inconsistent or meaningless names
• 3: Mostly follows conventions, some issues
• 5: Clear, consistent, idiomatic names

Code Organization (1-5)

• 1: Monolithic, hard to navigate
• 3: Some structure, could be cleaner
• 5: Well-organized, logical structure

Error Handling (1-5)

• 1: Missing or swallowed exceptions
• 3: Basic error handling present
• 5: Comprehensive, appropriate error handling

Documentation (1-5)

• 1: No documentation
• 3: Some documentation, inconsistent
• 5: Thorough, helpful documentation

Idiomaticity (1-5)

• 1: Non-idiomatic, smells like translated code
• 3: Mostly idiomatic, some foreign patterns
• 5: Fully idiomatic, natural code

Output Format

Save scores as a JSON file:

{
  "target_file_a.java": {
    "naming_conventions": 4,
    "code_organization": 3,
    "error_handling": 5,
    "documentation": 4,
    "idiomaticity": 4,
    "justification": "Good naming and error handling. Some methods could be extracted for better organization."
  }
}

Custom Rubric

If a custom rubric is provided, use its attributes and scoring criteria instead of the defaults. The rubric should define:

• Attribute names
• Score meanings for each level (1-5)
• Examples of good/bad code for each attribute

Evaluation Guidelines

What to Look For

Positive indicators:

• Consistent naming (camelCase for Java, snake_case for Python)
• Appropriate class/method sizes
• Single responsibility principle followed
• Meaningful comments (not obvious ones)
• Standard library usage over custom implementations
• Defensive programming practices

Negative indicators:

• Literal translations (COBOL-style Java)
• God classes or methods
• Magic numbers
• Excessive comments or no comments
• Reinvented wheels
• Swallowed exceptions
• Hardcoded values

Language-Specific Considerations

Java:

• Streams vs. for loops (both acceptable, but be consistent)
• Optional vs. null checks
• Record types for data classes
• Builder pattern for complex objects

Python:

• Type hints
• List comprehensions
• Context managers
• Pythonic idioms

Incremental Scoring

If a score file already exists, update it with new scores rather than replacing it entirely.