AI Security Scanner

A command-line tool that scans a source code file for security vulnerabilities using Google's Gemini API. Point it at a file, and it sends the code to Gemini 2.5 Flash with a security-review prompt, then prints the findings back to the terminal — sorted by severity and color-coded for quick scanning.

• AI-powered vulnerability analysis via the Gemini API (gemini-2.5-flash)
• Findings sorted by severity (Critical → High → Medium → Low)
• Color-coded terminal output per severity level (via colorama)
• Single-file, single-command usage — no config beyond an API key
• Works on any source file that can be read as text (not limited to Python)

• Language: Python
• AI model: Google Gemini (gemini-2.5-flash) via the google-genai SDK
• CLI/output: colorama for colored terminal text
• Config: python-dotenv for loading the API key from a .env file

.
├── scanner.py       # Main CLI: reads a file, queries Gemini, prints sorted/colored findings
├── vulnerable.py    # Sample file with intentionally vulnerable code, for testing the scanner
├── images/demo.png  # Screenshot used in this README
└── .env             # Local Gemini API key (not committed with a real key)

• Python 3.10+
• A Google Gemini API key (Google AI Studio)

git clone https://github.com/levibmackay/SecurityScanner.git
cd SecurityScanner

Mac/Linux:

python3 -m venv venv
source venv/bin/activate

Windows:

python -m venv venv
venv\Scripts\activate

pip install google-genai python-dotenv colorama

Create a .env file in the project root:

GOOGLE_API_KEY=your_api_key_here

scanner.py exits with an error if GOOGLE_API_KEY isn't set.

python scanner.py <path_to_file>

Example, using the included sample file:

python scanner.py vulnerable.py

Severity: Critical
Vulnerability: SQL Injection
Why: User input is directly concatenated into a SQL query.
Impact: Attackers can execute arbitrary SQL commands.
Fix: Use parameterized queries.
--------------------------------------------------
Severity: High
Vulnerability: Command Injection
Why: User input is passed directly to a shell command via os.system.
Impact: Attackers may execute arbitrary shell commands.
Fix: Avoid shell execution; validate and sanitize input.

vulnerable.py is a deliberately insecure file for exercising the scanner. It contains:

• SQL injection (unsanitized string interpolation into a query)
• Command injection (os.system with unsanitized input)
• Weak password hashing (MD5)
• Hardcoded credentials and an API secret in source

1. Reads the target file's contents as text.
2. Fills a fixed security-review prompt with that code and sends it to Gemini 2.5 Flash.
3. Splits the response into individual findings (separated by ---) and sorts them by severity, based on keyword matching in each block.
4. Prints each finding to the terminal, colored by severity (red for critical, light red for high, yellow for medium, green for low).

• Severity sorting and coloring rely on keyword matching against the response text, not structured output from the model — a finding whose Why/Impact text happens to mention a different severity word than its own Severity: line can be misclassified.
• No error handling for a missing/unreadable input file path; an invalid path raises an unhandled exception instead of a clean error message.
• The security-review prompt and model name are hardcoded in scanner.py; there is no way to point at a different model or customize the prompt without editing source.
• No automated test suite.

Last updated: July 22, 2026

Last reviewed: 2026-07-20 19:33 MDT

Last updated: 2026-08-11 06:03 MDT

Maintained by Levi Mackay