StyleShield AI

Overall rating of the submitted new source code
To download the generated report
Similarity score between previous and newly submitted source code
Generated Coding Fingerprint
Generated radar chat
Supporting WebMCP tools
Final result after analysing

Inspiration

We were inspired by a simple problem: How can we verify whether a programmer actually wrote a piece of code?

Traditional plagiarism and code-similarity tools mainly compare the code itself. But programmers have individual habits in the way they name variables, structure functions, use loops, format code, write comments, and manage complexity.

We wanted to turn these behavioral patterns into a unique Coding Fingerprint, similar to a digital identity for programming style. This idea led to StyleShield AI.

What it does

StyleShield AI analyzes a programmer's previous code samples and creates a Coding Fingerprint and Coding DNA based on their programming behavior.

When a new code submission is provided, StyleShield:

Extracts structural and behavioral coding features
Compares the new code with established coding patterns
Calculates a similarity score
Uses machine learning for prediction
Generates Coding DNA
Provides programming-style/personality insights
Produces an explainable verification result

For the WebMCP Challenge, we extended StyleShield with WebMCP tools such as analyze_code, get_coding_fingerprint, and explain_verification. This allows AI agents to directly access StyleShield's analysis capabilities through structured tool calls while keeping the human involved in the final verification decision.

How we built it

We built StyleShield using Python, Flask, JavaScript, SQLite, and machine-learning techniques.

The system extracts features such as variable naming, indentation, function structure, complexity, loops, comments, and AST-based characteristics. These features are combined to create a programmer's Coding Fingerprint and Coding DNA.

We then developed a Flask API that connects the frontend to the analysis modules.

For the WebMCP extension, we used document.modelContext.registerTool() to expose StyleShield functionality as structured WebMCP tools. These tools communicate with our Flask API and return structured JSON results that an AI agent can understand and use.

Challenges we ran into

One of our biggest challenges was connecting WebMCP with our existing Flask application while keeping the existing StyleShield functionality stable.

We faced issues with asynchronous JavaScript requests, API responses, and exposing our local Flask application through a Cloudflare Quick Tunnel. At one point, the tunnel returned an HTTP 530 error, which initially looked like an application failure.

After debugging, we identified the problem as tunnel connectivity rather than our StyleShield backend. We restarted the tunnel and successfully established communication between the WebMCP layer and our Flask API.

Another challenge was representing programming style accurately. A programmer's identity cannot be determined from one feature, so we combined multiple coding characteristics to create a more meaningful fingerprint.

Accomplishments that we're proud of

We are proud that we transformed an existing programmer-code analysis concept into a WebMCP-enabled AI-assisted verification system.

Our key accomplishments include:

Building a working Coding Fingerprint system
Generating unique Coding DNA
Combining multiple programming-style features
Integrating machine-learning-based prediction
Building an explainable verification workflow
Successfully registering multiple WebMCP tools
Connecting WebMCP directly to our Flask analysis backend
Successfully receiving structured analysis results through the WebMCP-connected application

Most importantly, we learned how an existing application can be extended so that AI agents can directly use its capabilities instead of only interacting with its visual interface.

What we learned

We learned that programming style contains many subtle behavioral patterns that can be useful for identity verification.

Technically, we learned how to build modular AI applications with Flask, connect frontend tools to backend APIs, handle asynchronous requests, work with structured JSON responses, and integrate WebMCP into an existing web application.

We also learned that good human-agent interaction is not about replacing human decisions. Instead, an AI agent can gather evidence, perform analysis, and explain results while the human remains responsible for the final decision.

What's next for StyleShield AI

We plan to make StyleShield more robust and useful in real-world environments.

Future improvements include:

More advanced behavioral and semantic code analysis
Larger and more diverse training datasets
Cross-language programmer fingerprinting
Stronger confidence and explainability mechanisms
Continuous learning from new verified code samples
Integration with coding assessment and educational platforms
Advanced AI-agent workflows using additional WebMCP tools
Programmer verification history and audit trails
Better protection against intentional style imitation

Our long-term goal is to make StyleShield a reliable programmer identity intelligence layer that can assist organizations, educators, and developers in understanding whether code is consistent with a programmer's established coding behavior.