Introduction

Completed

Agents powered by the GitHub Copilot harness use natural-language instructions to pursue goals, respond to users, and decide how to use their available knowledge. Instead of defining a topic path for each conversation, you describe the agent's job and the rules it should follow.

Effective instructions begin with a clear purpose and scope. They also define how the agent communicates, which information it should use, what it must avoid, and when it should refer a user to a person or another process. You then test those decisions in Preview and refine the instructions based on observed behavior.

In this module, you learn how to:

  • Create an agent powered by the GitHub Copilot harness.
  • Define an agent's purpose and scope.
  • Write instructions for behavior, tone, grounding, boundaries, and escalation.
  • Add file and website knowledge sources.
  • Configure agent details, a greeting, and suggested prompts.
  • Establish baseline behavior in Preview.
  • Refine instructions based on test results.

The module focuses on creating, grounding, testing, and refining an instruction-driven agent.