How can we automate the disclosure of what AI agent was used in a PR and the extent of code? Would be nice to also have an audit of prompts used, as that could also be considered “code”.