From Unknown to Understood: Navigating Codebases with GitHub Copilot

Engineers frequently switch teams, projects, and navigate complex codebases they didn't write. Understanding the interconnectedness of code can be time-consuming and challenging, especially under pressure. The author reframed AI as an investigative partner to trace execution paths and uncover insights efficiently. They initially used the Plan agent in Github Copilot to investigate code and document findings. The investigation results were then manually moved, so they incorporated Atlassian MCP to directly upload information to Confluence or Jira. This process was improved by a custom agent, "Super-Investigator," which integrated the best features. The Super-Investigator explores the codebase, documents findings, diagrams relevant information, and creates a Confluence page. It then summarizes findings and links to the Confluence page in the associated Jira ticket. The custom agent is more efficient and accurate than the original process because it stores more relevant data. Developing this tool took two weeks and provided faster, more detailed investigations and will continue to improve with usage. The author is excited to see its future capabilities and sharing it with their team.