For Microsoft Copilot Studio agents, effective prompting depends on giving the model strong guidance and continuously improving that guidance from real interactions.

C. Use clear and specific instructions in the prompts is correct because well-written prompts reduce ambiguity, improve intent interpretation, and make the agent’s behavior more reliable and consistent.

D. Regularly test and refine the prompts based on user input is also correct because prompt quality improves through iteration. Reviewing how users actually interact with the agent helps identify unclear wording, missing instructions, and failure cases.

Why the others are not correct:

A. Measure system resource usage during prompt processing is an operations/performance activity, not a prompt best practice.

B. Analyze the prompt length distribution may be interesting analytically, but it is not a primary best practice for improving agent interactions.

E. Track the duration of the average user session is a usage metric, not a prompt-engineering best practice.