agent-security-scanner

This capability utilizes AST (Abstract Syntax Tree) parsing via tree-sitter to analyze code for vulnerabilities across multiple programming languages. It allows for more accurate detection of security issues by understanding the code structure rather than relying solely on regex patterns, which are less reliable. When tree-sitter is unavailable, it falls back to regex, ensuring flexibility in various environments.

This capability implements taint analysis to track the flow of user input through the code, identifying potential security risks when user data reaches sensitive operations or functions. By analyzing the data flow, it can pinpoint where user input may compromise security, allowing developers to take corrective actions.

This capability analyzes detected vulnerabilities and provides automatic fix suggestions based on established security patterns and best practices. By leveraging a database of common vulnerabilities and their remedies, it assists developers in quickly addressing security issues in their code.

This capability scans and verifies package existence across multiple repositories, including npm, PyPI, and others, to detect hallucinated packages that may not exist. By cross-referencing a database of over 4.3 million packages, it ensures that developers are not misled by incorrect package names or versions.

This capability analyzes AI agent prompts to detect potential prompt injection attacks that could manipulate the agent's behavior. By examining the structure and content of prompts, it identifies suspicious patterns that may indicate an attack, allowing developers to secure their AI interactions.