AI Field Guide
Short, practical entries on getting the most out of AI tools. Browse by category, or search from the sidebar.
Claude Code commands
/goal/goal is a Claude Code command that lets you define a clear objective and have your AI agent work toward it autonomously, so it keeps going instead of stalling and waiting on you.
/insights/insights is a Claude Code command that reviews how you work with your AI agent and surfaces concrete, personalized ways to prompt and collaborate more effectively.
/loop/loop keeps your AI agent working on a repeating task in the background, prompting itself each round, so it babysits the boring jobs instead of you.
/remote-controlRemote control lets you launch and steer your AI coding agent from your phone, so you can keep long-running tasks moving while you're away from your computer.
Customizing your agent
Agent Skills
A Skill is a reusable text file that teaches your AI to do a recurring task your way, so it stops guessing and starts following your rules automatically.
Progressive disclosure
Structure a Skill so the AI reads a lean top file and only pulls in the detailed references when it needs them, which keeps your context small and your token bill down.
Building apps
Getting better output
Critique loop
Have a second, separate model review and critique your AI's work before you accept it, which reliably beats trusting the first draft or running lots of agents in parallel.
Evals
An eval turns 'does this output look good?' into a repeatable pass rate you can actually track, so you know if a prompt tweak or a new model made things better or worse.