Domain (04): (QRP) Prompt Crafting and Prompt Engineering

[MohamedRadwan-DevOps]

Domain (04): (QRP) Prompt Crafting and Prompt Engineering (9%)

Document Type: QRP (Quick Revision Pack)
Scope: This document provides a high signal, minimal noise revision pack for this domain, optimized for last mile review. It focuses on crisp bullet points, micro examples, and checklists that cover the most important concepts and exam relevant angles.

Fundamentals of prompt crafting

Prompt crafting/How context for the prompt is determined

• Copilot uses your prompt plus additional context such as the current file and (for chat) chat history.
• In IDE chat, you can improve relevance by explicitly referencing context (files, folders, symbols, workspace) using your editor’s context features (for example VS Code context mentions and workspace indexing).

Prompt crafting/Language options for prompting GitHub Copilot

• You can prompt using natural language questions/requests and include code snippets or comments to steer intent.
• For best control, mix: natural language goal + constraints + minimal code context (only what Copilot needs to act correctly).

Prompt crafting/Different parts of a prompt (high-signal structure)

• Goal: what you want (one sentence).
• Context: file/function/inputs/constraints (what Copilot must consider).
• Requirements: correctness, security, performance, style, libraries, test framework.
• Acceptance criteria: what “done” means (examples, edge cases, output format).

Prompt crafting/Role of prompting (why it matters)

• Prompting is how you shape the model’s behavior: scope, assumptions, output format, and tradeoffs.
• Good prompts reduce hallucinations by forcing Copilot to anchor on constraints and concrete checks.

Prompt crafting/Zero-shot vs few-shot prompting

• Zero-shot: give only the instruction (fast; good for simple tasks or when format doesn’t matter).
• Few-shot: include 1–3 small examples of the desired format/pattern (best when you need strict structure, consistent style, or a specific transformation).

Prompt crafting/How chat history is used with GitHub Copilot

• Chat can use prior messages as context; this helps continuity but can also cause drift.
• If answers degrade, restate constraints in one message or start a new thread focused on one task.

Prompt crafting/Best practices (Copilot-specific)

• Start general, then add constraints: goal → requirements → examples.
• Ask for alternatives and tradeoffs when needed (e.g., “Option A vs B with pros/cons”).
• Keep scope small: one function, one bug, one refactor step; iterate.
• Always validate: run tests, review for security and correctness before using output.

Prompt crafting/Quick check

• Micro example: “Write a function to parse JWT claims.” → improve prompt with: required library, validation rules, error handling, and 2 example inputs/outputs (few-shot if output shape must be exact).

Source:
Prompt engineering for GitHub Copilot Chat (GitHub Docs)
Best practices for using GitHub Copilot (GitHub Docs)
Manage context for AI (VS Code) (VS Code Docs)
How to write better prompts for GitHub Copilot (GitHub Blog)
Introduction to prompt engineering with GitHub Copilot (Microsoft Learn)

Fundamentals of prompt engineering

Prompt engineering/Principles (what “engineering” adds beyond basic prompting)

• Treat prompts as repeatable inputs: consistent structure, reusable templates, and clear constraints.
• Optimize for reliability: reduce ambiguity, force assumptions to be explicit, and demand acceptance criteria.
• Prefer iterative refinement: generate → critique → revise with tighter constraints.

Prompt engineering/Training methods (what to know for the exam)

• Prompting is not model training; it is steering a pretrained model at runtime using instructions and context.
• Few-shot examples act like “in-context guidance” (pattern-setting), not permanent learning.

Prompt engineering/Best practices (practical, testable habits)

• Use “start broad → add constraints → add examples” as the default loop.
• Specify what to avoid (unsafe APIs, forbidden libraries, performance limits) to reduce bad defaults.
• If output must be structured, explicitly request format (JSON schema, table columns, function signature) and include a sample (few-shot).
• Use context tools (workspace/file/symbol references) instead of pasting huge blobs of code.

Prompt engineering/Prompt process flow (high-level)

• Define goal + constraints → provide/attach the right context → send prompt → review output → validate (tests/security) → iterate with corrections.

Prompt engineering/Quick check

• Micro example: “Refactor this function” becomes engineered: “Refactor for readability, keep behavior identical, add unit tests for edge cases, and explain changes; do not change public API.”

Source:
Prompt engineering for GitHub Copilot Chat (GitHub Docs)
Copilot Chat cookbook (GitHub Docs)
Manage context for AI (VS Code) (VS Code Docs)
Introduction to prompt engineering with GitHub Copilot (Microsoft Learn)