Overview

Prompt engineering isn't clever wording. It's the discipline of getting an agent to do the right thing repeatably. Three ideas do most of the work, and everything else in this section builds on them.

The three that matter most

1. Structure the instruction — role, task, constraints, output shape, stated explicitly. Ambiguity is where agents improvise.
2. Manage the context — include what's needed to decide, exclude what isn't. More context is not better context.
3. Verify, don't assume — a verification loop is the difference between a demo and a tool.

Then: techniques and patterns

The techniques section covers decomposition, tool-use prompting, and evaluation. The patterns cookbook shows these composed into real workflows — code review, debugging — the same ones our skills encode.

The throughline

Once a prompt reliably produces the outcome you want, it stops being a prompt and becomes an Agent Skill: packaged, named, and loaded automatically whenever the task comes up. That handoff — from prompt to skill — is the point of everything here.