GH-300: Cheatsheet
Exam Day Reference
Review this page 5 minutes before the exam.
Responsible AI Core (HVA)
- Human-in-the-loop — Always validate AI output before using it
- Validate — Test, read, and review every suggestion
- Accountable — The developer is responsible, not the AI
Copilot Plans at a Glance
| Feature | Individual | Business | Enterprise |
|---|---|---|---|
| IDE suggestions & chat | ✅ | ✅ | ✅ |
| Org policy management | ❌ | ✅ | ✅ |
| Content exclusions | ❌ | ✅ | ✅ |
| Audit logs | ❌ | ✅ | ✅ |
| Knowledge Bases | ❌ | ❌ | ✅ |
| Copilot on GitHub.com | ❌ | ❌ | ✅ |
Mode Quick Reference
| Mode / Interface | Best For |
|---|---|
| Inline suggestions | Code as you type in IDE |
| Copilot Chat (IDE) | Questions, refactors, explanations |
| Agent Mode | Autonomous multi-step tasks |
| Edit Mode | Cross-file edits, review each change |
| Plan Mode | Preview Copilot's plan before execution |
| Copilot CLI | Terminal command help |
| Copilot on GitHub.com | PR summaries, code review (Enterprise) |
Chat Commands
/explain— Explain selected code/fix— Suggest a fix for selected code/tests— Generate unit tests for selected code/doc— Generate documentation for selected code@workspace— Search across the entire project
Prompt Engineering (GCSF)
- Goal — What to do
- Context — Why/where
- Source — What data/files to reference
- Format — Output style, types, error handling
Few-shot: Provide 1-3 examples before asking Zero-shot: Just describe the task Prompt file: .github/copilot-instructions.md for workspace-wide conventions
Safeguards Quick Reference
| Safeguard | What it does |
|---|---|
| Content exclusions | Prevents Copilot using specific files as context |
| Duplication detection | Filters suggestions matching public GitHub code |
| Security warnings | Flags suggestions with vulnerability patterns |
Quick Decision Rules
Autonomous multi-step task? → Agent Mode
Cross-file refactor, want to review each change? → Edit Mode
Need Copilot to use external service data? → MCP (Model Context Protocol)
Need persistent org-wide Copilot instructions? → .github/copilot-instructions.md
Org wants to block a sensitive directory from Copilot? → Content exclusions (requires Business/Enterprise)
Suggestion matches public open source code? → Duplication detection enabled — suggestion filtered
AI output seems wrong / hallucination? → Validate output independently; refine prompt with more context
Terminology Check
- Hallucination: Confident but incorrect AI output (wrong API names, invented functions)
- Zero-shot prompting: Task request with no examples — model uses training knowledge alone
- Few-shot prompting: Providing examples to guide the model's pattern
- Duplication detection: Filter for suggestions matching public GitHub repos verbatim
- Content exclusions: Admin config preventing specific files from being used as Copilot context
- MCP: Model Context Protocol — connects Copilot to external data sources at runtime
- Agent Mode: Copilot autonomously runs multi-step tasks including terminal commands
- Edit Mode: Copilot proposes cross-file edits; user reviews each change individually
- Prompt file:
.github/copilot-instructions.md— persistent workspace instructions for Copilot