Axiom – blocks AI agents from inventing limits your code never defined (github.com)

Axiom has introduced a new tool designed to enhance safety in AI coding agents, specifically integrating as a pre-write guardrail within Claude Code and Cursor environments. This tool operates with zero dependencies, requiring no API keys or tokens, and blocks potentially harmful code before execution based on regex and AST-pattern detection. While Axiom effectively addresses common coding violations, it is important to note that it may be bypassed by determined users through simple obfuscation techniques, highlighting the need for additional safeguards like container isolation and robust code reviews. The significance of Axiom lies in its ability to enforce coding standards and prevent accidental breaches of coding practices that can undermine application security, particularly in areas like authentication checks and sensitive data handling. It utilizes a three-state logic system to determine code validity, allowing AI agents to ask clarifying questions rather than silently executing harmful code. This approach not only aids developers in adhering to best practices but also fosters a culture of safety within AI-assisted coding, making it a crucial addition to the toolkit of AI/ML developers concerned with security and compliance.