AI coding agent harness that scans, scores, and systematically improves codebase quality — turning slop code into well-engineered, beautiful code.
Desloppify is an AI coding agent harness for codebase quality governance. It employs a dual-track evaluation system — mechanical detection and LLM subjective review — to quantitatively assess codebases across multiple dimensions. The core workflow follows scan → score → review → triage → execute → rescan, progressively advancing fix tasks via the next command, working in conjunction with AI coding agents like Claude, Cursor, and Copilot for actual code improvements.
It detects objective issues including dead code, duplication, complexity, and test coverage gaps, while also evaluating subjective dimensions such as naming quality, abstraction levels, and error handling patterns. The scoring system uses strict/lenient dual scores with anti-gaming design — wontfix items widen the strict/lenient gap, and re-review may lower scores. State is persisted in .desloppify/ for cross-session incremental technical debt reduction.
Supports 29 programming languages with full plugin depth for TypeScript, Python, C#, C++, Dart, GDScript, Go, and Rust, plus generic linter + tree-sitter coverage for 18+ additional languages. C++ uses compile_commands.json as its primary analysis path. Monorepo-friendly with per---path and --lang scanning. Scorecard badges available at 98+ scores.
Built on Python 3.11+, distributed via PyPI (pip install "desloppify[full]"), under OSNL-0.2 license (free for personal/internal use, free for open-source companies). CLI entry point is desloppify.cli:main, package structured by language sub-packages, built with setuptools, using Ruff + mypy for quality control, pytest covering all language sub-package tests.
Installation & Quick Start
pip install --upgrade "desloppify[full]"
desloppify update-skill claude # options: claude, cursor, codex, copilot, droid, windsurf, gemini
desloppify scan --path .
desloppify next
Monorepo Usage
desloppify --lang typescript scan --path ./frontend
desloppify --lang python scan --path ./backend
Key Commands: scan (scan & score), next (execute next task, core loop command), backlog (view all pending), plan (reorder priorities), exclude (exclude directories), update-skill (install AI agent workflow guides).
Unconfirmed: Specific LLM model and configuration for review phase; resolve command details; complete list of 29 languages; Scorecard badge format and hosting; existence of standalone configuration file mechanism.