Agent-trace: A standard format for tracing AI-generated code (github.com)

Agent Trace has been introduced as an open specification for tracing AI-generated code, providing a vendor-neutral format to accurately record contributions from both AI and human authors in version-controlled codebases. This initiative addresses the growing need for clear attribution in environments where AI assistance is increasingly prevalent, allowing developers and stakeholders to distinguish between AI contributions and those made by humans. Its significance lies in enhancing transparency, facilitating interoperability between tools, and supporting granular attribution at the file and line level. Key features of Agent Trace include an extensible metadata framework for adding vendor-specific data and a human-readable format that enables easy access to attribution details without specialized tools. It covers multiple version control systems (VCS) and enables tracking of code ownership through detailed "Trace Records," which include specific contributions and the context of the conversations associated with them. While it focuses on the attribution aspect and does not address legal ownership or quality assessment, Agent Trace sets the groundwork for a standardized approach to documenting AI-enabled coding activities, fostering clearer collaboration and understanding in the rapidly evolving AI/ML landscape.