Skip to content

README.md

Latest commit

 

History

History
40 lines (31 loc) · 2.36 KB

README.md

File metadata and controls

40 lines (31 loc) · 2.36 KB

Overview

"Code Slob" refers to the subtle technical debt, unnecessary verbosity, and over-complexity often introduced by AI coding agents or rapid development. While functional, "slob" code is harder to maintain and prone to future regressions. The Code Slob Cleanup project builds an automated toolchain—packaged as a coding agent skill—to identify, refactor, and rigorously verify Python code to remove this debt.

Currently, the project only supports python codebases.

Installation

Linux/Mac Auto Install/Uninstall/Update for Gemini/Claude/Codex:

curl -LsSf https://raw.githubusercontent.com/Jazz23/CodeSlobCleanup/refs/heads/main/install.sh | sh

Gemini CLI:

gemini extensions install https://github.com/Jazz23/CodeSlobCleanup

How to use

Click here for a guide on how to use Code Slob Cleanup.

Core Objectives

  1. Automated Identification: Detect refactoring targets using static analysis metrics (Cyclomatic Complexity, LoC, nesting depth) and semantic analysis via the agent to identify redundant logic.
  2. Safe Refactoring: Employs your agent to perform code transformations such as function decomposition, visibility enforcement (Converting public to private), and dead code removal.
  3. Rigorous Verification: Ensure behavioral equivalence using Property-Based Testing: Leveraging Hypothesis to verify Original(input) == Transformed(input) across thousands of inputs.
  4. Autonomous Self-Correction: Integrate verification feedback directly into the refactoring loop, allowing the agent to "fix its own fixes" based on counter-examples found during testing.

Architecture

The system follows an iterative loop:

  • Scanner: Traverses the codebase and finds code slob.
  • Refactor Agent: Analyzes flagged code and generates a cleaned-up version.
  • Verifier: An isolated orchestrator executes Hypothesis tests, and benchmarks performance.
  • Feedback Loop: If verification fails, the failing test cases are fed back to your agent for a retry.
  • Committer: Applies the change if verification passes and performance is not regressed.

Technologies

  • Language: Python 3.14+
  • Package Management: uv (PEP 723 for standalone scripts)
  • Testing: Hypothesis (Property-based), Pytest
  • Benchmarking: Matplotlib/Numpy for performance comparison