I Spent My Weekends Building an AI Debugger That Understands Your Code

The real problem: In my daily work, I found myself repeatedly spending too much time gathering runtime context just to explain bugs to AI tools like GitHub Copilot. Manually copying variables, tracing paths, and filtering noise is bit time taking.

So I built Co Debugger AI(VSCode Extension): Smart algorithms that instantly generate attachable, intelligent debugging context using Symbolic Analysis, Path Sensitivity Analyzer, and Smart Filtering Algorithm from your runtime state along with the runtime variables.

What it does: Transforms a runtime debugging session into AI-consumable context in one click.

Example - The Workflow

Before: Hit breakpoint Manually copy variable values Screenshot call stack Type context into Copilot chat Lose focus, forget details Get generic AI response

With Co Debugger AI: Hit breakpoint → Click status bar Select context type (10 options) File auto-opens in side panel Attach to Copilot → Ask intelligent questions

How smart analysis makes context AI-ready: • Path Sensitivity Analysis = Shows variable relationships across execution paths • Symbolic Execution Engine = Computes "what-if" branches without running code • Smart Filtering Algorithm = Identifies only the relevant variables for your case • AI-Optimized Formatting = Structures context for maximum AI comprehension

Perfect for AI assistance: Ask Copilot: "Why is my payment processing failing intermittently?" — with full context.

Yes, runtime debugging helps to solve a lot of bugs. I was so annoyed with buggy codes generated by AI so at the end I actually developed something not sure similar to your or not:both AI coder and AI debugger two-in-one, a fork of Roo-Code. So it (Zentara-Code) can code and debug it code at the same time in runtime. It automatically set breakpoints, do stack tracing, inspect stack frame variables. As it leverages Debug Adapter Protocol (DAP), it works for any language in VSCode. https://github.com/Zentar-Ai/Zentara-Code

This is an very exciting idea -- combining current symexe engines and CFG/dataflow analyzers with LLMs is indeed a promising topic.

Can't wait to read the code!