Using AI agents without clear guidance can feel like walking in the dark. One prompt works well. The next one takes you somewhere completely different. As agents start doing more work inside our codebases, how do we give them the context and structure they need?

I’ve been using skills and playbooks to give agents more structure. Skills give an agent reusable instructions and context for a specific kind of work. Playbooks lay out a repeatable process for tasks like building a feature, reviewing a pull request, or writing tests. Instead of explaining the same things over and over, the agent can load the right guidance when it needs it. This approach can work across tools like Claude Code, Codex, and Cursor.

In this session, I will show how I structure skills and playbooks, decide what belongs in each, and keep the context focused. We will also look at common problems like conflicting instructions, stale guidance, and agents choosing the wrong workflow. You will leave with practical ideas you can use to move beyond one-off prompts and give agents a more repeatable way to work in your codebase.