Two users, one CLI: people and agents

Building an internal command-line interface (CLI) for human developers revealed design flaws when AI agents also began using the tool. Interactive prompts and implicit states, which humans handle easily, caused agents to fail. The solution involved creating programmatic equivalents for every interactive input, making state explicit, and providing structured context files for agents to consume. This process of designing for agents ultimately resulted in a more robust, scriptable, and testable tool that benefits both human and automated users.