5.3. Reflection Checkpoint

Key Takeaways

• A good prompt does two things: states intent clearly and supplies relevant context. Every tactic serves one of those; length is not a goal, reduced ambiguity is.
• Copilot determines context from your editor state — open files, selection, neighboring tabs — not just typed words. You craft context by managing your workspace.
• Zero-shot vs. few-shot is about whether the answer's form needs demonstrating. Few-shot shapes specific patterns; zero-shot is leaner for common tasks. "Few-shot always wins" is false.
• Prompt engineering rests on principles (clarity, relevance, specificity, decomposition, iteration) that explain why tactics work — and on a process flow where Chat history feeds each new request, making follow-ups work but also allowing stale context to mislead.

Connecting Forward

Phases 1-5 built the foundation: what Copilot is, what it can do, how to use it responsibly, how it works, and how to prompt it well. Phase 6 puts it to work — using Copilot to actually improve developer productivity across generation, refactoring, documentation, testing, and security. Good prompting is the engine; productivity is the payoff.

Self-Check Questions

• Rewrite a vague prompt ("fix this function") into a strong one, and name each ambiguity you resolved.
• A developer's suggestions are generic despite a clear instruction. Walk through the context levers (open files, selection, instructions files) you'd adjust before blaming the model.
• Explain why few-shot prompting is a tool for shaping output form rather than a universal upgrade, and give one case where zero-shot is clearly better.