Logging for AI-assisted development
When a prototype uses data or calls tools, plan how its actions will be recorded and reviewed.
Read implementation guidanceDefine the logging requirements
Record the tool invoked, the initiating user or agent, the relevant permissions, and the result. Include approvals and exceptions where they are part of the workflow.
Decide which data belongs in the record. Avoid copying credentials or unnecessary sensitive information into logs.
Review a complete workflow
Follow a request through the prototype and check whether the records explain what happened. Review who can read those records and how an error would be investigated.
For systems that use MCP servers, include tool calls and their permission checks in the review.
Documentation
See TN-Proto implementation guidance for the protocol work, and our overview of AI audit trails for the records to consider.