🧬 The Self-Healing AI: A Multi-Agent Debugger

Autonomous Hacks 2026 Final Project · v1.0.0 — Genesis: Autonomous Self-Healing Debugger

"Software breaks. Usually, humans have to fix it. Today, the software fixes itself."

---

🌟 Vision

We are building the immune system for software.

Debugging is a major bottleneck in modern development—it requires human intuition to read logs, understand context, and write patches. The Self-Healing AI automates this entire loop using a state-aware Multi-Agent System (MAS) that behaves like a digital hospital for broken code:

• 🦠 Detects the illness (bug)
• 🩺 Diagnoses the root cause
• 🛠️ Performs surgery (code fix)
• ❤️‍🩹 Validates recovery (verification)

---

⚙️ Architecture

A sequential, state-aware multi-agent workflow where agents collaborate through shared artifacts.

Crash
  ↓
Trace Log (trace.log)
  ↓
Scanner Agent → Diagnosis (diagnosis.txt)
  ↓
Fixer Agent → Patched Code (.py)
  ↓
Validator Agent → Validation Report

---

🧠 Agents

🔍 Scanner Agent — The Doctor

• Role: Observability & Diagnosis
• What it does: Parses raw Python tracebacks and converts them into human-readable reasoning
• Output: diagnosis.txt

🛠️ Fixer Agent — The Surgeon

• Role: Active Repair
• What it does: Reads the diagnosis and rewrites the faulty source code (not suggestions)
• Output: Patched .py files

✅ Validator Agent — Quality Assurance

• Role: Verification
• What it does: Re-runs the failing workflow to ensure the fix works and introduces no regressions
• Output: Validation success / failure

---

🚀 Quickstart

🔧 Requirements

• Python 3.10+
• Google Gemini API Key
• Google Agent Development Kit (ADK) environment configured

⚠️ This project requires Gemini + ADK. Ensure your environment can run Gemini-powered agents.

---

🛠️ Setup

git clone https://github.com/JeetP1711/RagBooms-AutonomousHacks-Phase-01.git
cd RagBooms-AutonomousHacks-Phase-01
pip install -r requirements.txt

Create a .env file:

GOOGLE_API_KEY=your_api_key_here

---

▶️ Run the Demo (4 Steps)

1) Create the Crash

python 1_create_error_log.py

2) Diagnose

python 2_run_debugger.py

📄 Generates diagnosis.txt

3) Repair

python 3_run_fixer.py

🛠️ Modifies broken_researcher/agent.py in real time

4) Validate

python 4_run_validator.py

✅ VALIDATION SUCCESS

---

🏆 Hackathon Alignment

Problem Statement #2 — Multi-Agent Debugger Workflow

• Autonomous agent collaboration
• Real-time code repair
• End-to-end validation

---

⚙️ Tech Stack

• Python 3.10+

• Google Gemini API

• Google Agent Development Kit (ADK)

• Tool-enabled agents for:

  • Log parsing
  • File editing
  • Execution control

---

⚠️ Limitations

• Direct file overwrites (no PR flow yet)
• Local validation (no sandbox)
• Single-fault repair per cycle
• Requires external API access

---

🔮 Roadmap

• Human-in-the-loop approvals
• GitHub PR-based fixes
• Docker sandboxed validation
• Security & performance agents
• Multi-fault debugging

---

📦 Release

• Current: v1.0.0
• Name: 🧬 Genesis: Autonomous Self-Healing Debugger
• Status: Experimental / Proof-of-Concept

---

🏷️ Topics

self-healing-ai · multi-agent-system · autonomous-debugging · ai-debugger · software-repair · python · google-gemini · agent-development-kit

---

❤️ Acknowledgements

Built with ❤️ using Google Gemini and the Google Agent Development Kit (ADK) for Autonomous Hacks 2026.