Track
Reasoning Agents (Azure AI Foundry)
Project Name
Zero-Shield CLI
GitHub Username
jerisadeumai
Repository URL
https://github.com/jerisadeumai/zero-shield-cli
Project Description
Zero-Shield CLI is an AI-native security orchestrator designed for rapid cloud threat remediation. Built on the OODA loop (Observe-Orient-Decide-Act) framework, it utilizes GitHub Models (GPT-4o) and AWS Boto3 to bridge the gap between detection and action. The agent reasons through conversational input to identify compromised AWS resources, extract metadata, and execute precise quarantine protocols—such as security group isolation and ingress revocation—in real-time. By automating the extraction of instance IDs and regions from natural language, it reduces Mean Time to Repair (MTTR) from minutes to seconds, providing a hardened, agentic defense layer for cloud infrastructure.
Demo Video or Screenshots
📺 Demo Video
Watch the Zero-Shield CLI Demo on YouTube
⚡ Execution Screenshot
(Above: The Zero-Shield agent executing the quarantine protocol on a compromised instance via Boto3.)
Primary Programming Language
Python
Key Technologies Used
Key Technologies Used
- LLM: GitHub Models (GPT-4o)
- Cloud Provider: AWS (EC2, Boto3 SDK)
- Framework: OODA-loop Reasoning Engine
- Configuration: python-dotenv
Submission Type
Individual
Team Members
No response
Submission Requirements
Quick Setup Summary
Quick Setup Summary
- Clone & Install:
git clone https://github.com/jerisadeumai/zero-shield-cli && pip install -r requirements.txt
- Configure: Add your
GITHUB_TOKEN to a .env file.
- Launch: Run
python3 zero_shield_cli.py.
- Command: Input natural language like "Show running instances in us-east-1" or "Isolate instance i-12345."
Technical Highlights
I am most proud of the multi-stage OODA-loop reasoning engine. Instead of a simple pass-through to an LLM, the agent performs a distinct "Orient" phase where it validates extracted Instance IDs and Regions against real-time Boto3 metadata before moving to the "Act" phase. This prevents common hallucinations regarding infrastructure state and ensures that destructive security actions (like quarantining an instance) are only performed on verified targets.
Challenges & Learnings
The biggest challenge was the high-fidelity extraction of AWS metadata from unstructured prompts. I learned that by structuring the system prompt to explicitly follow the OODA framework, the model became significantly more reliable at "deciding" when it had sufficient information to act and when it needed to "observe" more data first.
Contact Information
contact.jerisadeumai.whenever652@slmails.com
Country/Region
India
Track
Reasoning Agents (Azure AI Foundry)
Project Name
Zero-Shield CLI
GitHub Username
jerisadeumai
Repository URL
https://github.com/jerisadeumai/zero-shield-cli
Project Description
Zero-Shield CLI is an AI-native security orchestrator designed for rapid cloud threat remediation. Built on the OODA loop (Observe-Orient-Decide-Act) framework, it utilizes GitHub Models (GPT-4o) and AWS Boto3 to bridge the gap between detection and action. The agent reasons through conversational input to identify compromised AWS resources, extract metadata, and execute precise quarantine protocols—such as security group isolation and ingress revocation—in real-time. By automating the extraction of instance IDs and regions from natural language, it reduces Mean Time to Repair (MTTR) from minutes to seconds, providing a hardened, agentic defense layer for cloud infrastructure.
Demo Video or Screenshots
📺 Demo Video
Watch the Zero-Shield CLI Demo on YouTube
⚡ Execution Screenshot
(Above: The Zero-Shield agent executing the quarantine protocol on a compromised instance via Boto3.)
Primary Programming Language
Python
Key Technologies Used
Key Technologies Used
Submission Type
Individual
Team Members
No response
Submission Requirements
Quick Setup Summary
Quick Setup Summary
git clone https://github.com/jerisadeumai/zero-shield-cli && pip install -r requirements.txtGITHUB_TOKENto a.envfile.python3 zero_shield_cli.py.Technical Highlights
I am most proud of the multi-stage OODA-loop reasoning engine. Instead of a simple pass-through to an LLM, the agent performs a distinct "Orient" phase where it validates extracted Instance IDs and Regions against real-time Boto3 metadata before moving to the "Act" phase. This prevents common hallucinations regarding infrastructure state and ensures that destructive security actions (like quarantining an instance) are only performed on verified targets.
Challenges & Learnings
The biggest challenge was the high-fidelity extraction of AWS metadata from unstructured prompts. I learned that by structuring the system prompt to explicitly follow the OODA framework, the model became significantly more reliable at "deciding" when it had sufficient information to act and when it needed to "observe" more data first.
Contact Information
contact.jerisadeumai.whenever652@slmails.com
Country/Region
India