Hi everyone!
Like many of you, I use AI assistants (Claude, Cursor) daily. But I noticed a problem: AI often suggests changes without understanding the full architecture. It might suggest deleting a file that seems unused but is actually dynamically imported, or it doesn't see the "blast radius" of a refactoring change.
So I built Code Health System — an open-source toolkit that acts as a context layer for AI agents.
(Full disclosure: I'm a young developer, and this project was built with the heavy assistance of AI/Cline. I'm trying to learn and create something useful using modern AI-native workflows.)
🚀 What makes it unique?
It’s not just a linter; it’s a safety layer for your AI assistant.
🏝️ Dead Island Finder: Instead of finding single unused functions (which creates noise), it finds clusters of files that form isolated "islands" of dead code. Safe to delete the whole module!
💥 Blast Radius Prediction: Before changing auth.py, ask: "What happens if I change this?". It predicts the cascade of errors across the project.
🤖 MCP Integration (For AI): This is the main goal. It runs as a local server. You just ask questions like ask("Can I delete services/old.py?") and it checks dependencies, git history, and safety.
⏳Sequence tool - you can enable up to 10 tools sequentially and get a mini-report, instead of running LLM 10 requests at a time.
❓ FAQ (Anticipating your questions)
Q: How is this different from vulture or pylint*?* A: vulture finds unused variables/code items. Code Health System finds architectural patterns. vulture says "this function is unused", but it might be wrong (dynamic import). My tool analyzes entry points and dependency graphs to say "this whole folder is an isolated island that no one calls". It's safer.
Q: What is "MCP"? A: You already know what "MСP" is. It's a convenient AI tool used in Claude Desktop, Cursor, and Windsurf. This program was written in vscode and tested in Cline.
Q: Is it safe to run? A: Yes. It runs locally on your machine. It doesn't send your code to the cloud. It just analyzes the AST and graph locally.
💬 Feedback needed!
I’m a young developer, and this is my first serious open-source release. I’m very interested to know: Is this tool actually useful to people? Does the concept of "Context for AI" make sense?
I’m looking for feedback on the architecture, code quality, and whether I should continue developing this. If you have a minute, please check the repo.
GitHub: https://github.com/atm0sph3re/code-health-system
PyPI: pip install code-health-system
P.S. > "Built with AI assistance by a junior developer. Feedback welcome!"
Top comments (0)