Skip to main content
Back

Desloppify Turns Messy AI Coding Agents Into Senior Engineers

A Python agent harness gives coding assistants a scoring system and fix loop to systematically improve code quality across 29 languages.

Pro

Repo, Full Access
  • Desloppify is a Python agent harness that scores and iteratively improves codebase quality.
  • Combines mechanical detectors (dead code, complexity) with LLM review of naming, abstractions, and module boundaries.
  • Persistent state in .desloppify/ lets agents chip away across sessions rather than one-shot fixes.
  • Supports 29 languages, with deep plugin support for TypeScript, Python, C#, C++, Go, and Rust.
  • Works with Claude Code, Cursor, Codex, Copilot, Droid, Windsurf, Gemini, and Rovodev via skill installers.
  • Available on PyPI, requires Python 3.11+, free for individuals and open-source companies.

A new open-source project called desloppify tries to turn AI coding agents from fast-but-messy contributors into something closer to a diligent senior engineer. The tool wraps your agent in a harness that scans, scores, and iteratively fixes a codebase until it hits a quality target you define, with anti-gaming logic baked into the score itself.

Desloppify gives an AI coding agent the tools to identify, understand, and systematically improve codebase quality. It combines mechanical detection (dead code, duplication, complexity) with subjective LLM review (naming, abstractions, module boundaries), then works through a prioritized fix loop. State persists across scans, so the agent chips away at technical debt over many sessions rather than doing one shallow pass.

A north star score your agent can chase

At the center sits a single score the agent optimizes against. According to the project, a score above 98 should correlate with a codebase a seasoned engineer would call beautiful. That score also generates a badge you can drop on your README.

Desloppify scorecard dashboard showing a score of 87.3 with breakdowns for file health, code quality, API coherence, security, and test health

Under the hood, the workflow is a six-step loop: scan, score, review, triage, execute, rescan. Scan runs mechanical detectors across your codebase for dead code, duplication, complexity, test coverage gaps, naming issues, and more. Review uses an LLM to assess subjective quality dimensions like naming, abstractions, error handling patterns, and module boundaries. Triage clusters findings into an ordered execution queue, and Execute runs the fix loop: