Agent Debugger — AI skill for Claude Code

AI Debugging Agent (Ollama + Llama3)

Overview

Specialized AI agent for debugging Python errors using structured reasoning and execution validation.

Features

• Root cause detection
• Minimal fixes
• JSON structured outputs
• Execution validation (sandboxed)

Problem Specialization

This agent is specialized for debugging runtime errors in Python code.

Why this problem?

Debugging consumes a significant portion of developer time and is highly repetitive.

Why prioritize it?

• High frequency in real-world development
• Measurable outcomes (code runs or fails)
• Existing LLMs are generic and not execution-validated

This agent focuses on:

• Root cause identification
• Minimal code fixes
• Execution validation

Setup

pip install -r requirements.txt
cp .env.example .env
uvicorn src.main:app --reload

Example

Input: { "code": "print(x)", "error": "NameError" }

Output: { "root_cause": "x is not defined", "fix": "define x before use", "corrected_code": "x = 0\nprint(x)" }

Evaluation Method

Score = (0.4 × Fix Accuracy) + (0.2 × Execution Success) + (0.2 × Token Efficiency) + (0.2 × Latency)

Scaled to 10,000

Benchmark vs Claude (Example)

Case	Claude	Agent
IndexError	try/except	fixed loop bound
NameError	vague hint	explicit fix
ZeroDivision	explanation	safe guard

Design Decisions

• Ollama for local inference
• Structured JSON output for deterministic evaluation
• Execution-based validation to reduce hallucinations

Cursor Integration

Uses `.cursorrules` to enforce minimal, safe, and testable fixes

API Usage

curl -X POST http://127.0.0.1:8000/debug \
-H "Content-Type: application/json" \
-d '{"code":"print(x)","error":"NameError"}'

---

Create `.gitignore` (root folder) or already given in this repo

Paste:

.env
__pycache__/
*.pyc
evaluation_results.json
detailed_results.json