Summary

  • 1 minute

In this module, you explored how instructions, knowledge sources, and testing shape the behavior of an agent powered by the GitHub Copilot harness. You learned how to:

  • Define an agent's audience, purpose, supported outcomes, and boundaries.
  • Write instructions for role, behavior, grounding, tone, clarification, and escalation.
  • Add file and website knowledge sources.
  • Align agent details, a greeting, and suggested prompts with the agent's scope.
  • Establish baseline behavior in Preview.
  • Refine instructions and rerun tests after observing a failure.

Learn more