code-smell-detector

Detect code smells in ML pipelines by analyzing training scripts for data leakage, magic hyperparameters, and randomness without seed. This identifies hidden design issues specific to machine learning workflows that standard static analysis often misses.

Common code smells in LLM-integrated applications include god prompts, generic roles, and hallucination feedback loops. Detecting these anti-patterns early helps cut technical debt and prevents underlying design problems in large language model workflows.

Find prioritized refactoring suggestions for technical debt across C#, Java, Python, C, C++, JS/TS, and SQL. The analysis provides severity-rated findings with concrete refactor guidance and BAD/SAFE code examples for each supported language.

Code smell detection works with standard static analysis tools by providing integration guidance for SonarQube, MLScent, and jscpd. It cross-references findings to related quality skills, extending traditional static analysis to cover ML pipelines and LLM workflows.

The best way to identify data leakage and magic hyperparameters is to run code smell detection during code review. This catches ML-specific design issues early with severity-rated findings and concrete refactoring suggestions before they become expensive to fix in production.

During code review, look for classic Fowler catalog anti-patterns like long methods and data clumps, alongside modern 2026-era additions for ML and LLM workflows. Detecting these anti-patterns generates prioritized refactoring suggestions to reduce technical debt.