Skip to content

Overview

The Agent Framework covers what agents are made of — the agent loop, tools, bots, skills, files, memory, and proactive triggers. Seven building blocks. Once you understand them, you can build an agent.

But understanding the blocks is not the same as knowing how to use them well. That is what the Agent Design Guide is for.

Think of the Agent Framework as understanding the materials — wood, nails, hinges, paint. The Agent Design Guide is carpentry — how to use those materials to build something that actually works.

The craft of agent design comes down to one question: what goes where? Every agent has the same building blocks available. What separates a well-designed agent from a mediocre one is how the designer distributes knowledge and behavior across those blocks.

An agent’s configuration has four layers:

System prompt — designer-authored, always present. This is where you define the agent’s identity, give it the context it needs, and set the instructions that govern how it operates.

Dynamic instructions — just-in-time, loaded or triggered as needed:

  • Skills — On-demand procedures, loaded when the task calls for them.
  • Schedules — Time-triggered actions.
  • Heartbeat — Ambient awareness between conversations.

Each of these is covered in its own page. The guide closes with The Design Flow — how to put these pieces together when designing a new agent.

Next: System Prompt →