GitHub Copilot (GH-300) Exam Prep
Prepare for Exam GH-300: use Copilot in the IDE, CLI and on GitHub, craft prompts, manage policies and content exclusions, and apply responsible AI, domain by domain.
What you will learn
- Describe the GH-300 exam and apply responsible AI principles, including validating Copilot output.
- Enable and use Copilot in the IDE: inline suggestions, Copilot Chat modes, commands and feedback.
- Use Copilot CLI, agent mode, MCP servers and the Copilot coding agent safely.
- Use Copilot on GitHub.com: code review, pull request summaries, Spaces, Spark, instructions and prompt files.
- Explain how Copilot builds prompts, filters and processes suggestions, and where LLMs fall short.
- Craft effective prompts and context with zero-shot, few-shot and chat history techniques.
- Use Copilot to generate, refactor, document and test code and to suggest security and performance fixes.
- Manage organization policies, audit logs, seats, content exclusions and the public code filter, and troubleshoot.
Modules
- The GH-300 exam and responsible AI
- Copilot in the IDE: suggestions and Chat
- Copilot CLI, agent mode, MCP and the coding agent
- Copilot on GitHub.com: reviews, Spaces, Spark and custom instructions
- Copilot data flow and architecture
- Prompt engineering and context crafting
- Developer productivity, testing and security
- Administration, privacy and safeguards
Course facts
- Level
- Exam preparation
- Estimated study time
- about 2 hours
- Final assessment
- 44 questions · Pass mark 70%
- Available in
- English, Français
- Price
- Free. Optional verifiable certificate: $5
- Last updated
- Published by
- AMAADOR ACADEMY
Who this course is for
Developers, DevOps engineers, team leads and GitHub administrators who use or manage GitHub Copilot and want to pass Exam GH-300.
Prerequisites
GitHub fundamentals (repositories, branches, pull requests) and working experience with at least one programming language. Hands-on access to Copilot is strongly recommended.
Course content
1. The GH-300 exam and responsible AI
The GitHub Copilot certification is earned by passing Exam GH-300. It is now delivered through Microsoft Learn and scheduled with Pearson VUE, online or at a test centre. The facts below were checked on Microsoft Learn (certification page and GH-300 study guide) in November 2026.
| Item | Confirmed fact |
|---|---|
| Exam code | GH-300, skills measured as of 7 August 2026 |
| Time | 100 minutes, proctored |
| Passing score | 700 or greater (scaled score) |
| Languages | English, Spanish, Portuguese (Brazil), Korean, Japanese |
| Price | Set by the country where the exam is proctored |
| Practice | Free official practice assessment and exam sandbox |
| Retake | First retake possible after 24 hours |
Domains and weights
- Use GitHub Copilot responsibly — 15–20%
- Use GitHub Copilot features — 25–30%
- Understand GitHub Copilot data and architecture — 10–15%
- Apply prompt engineering and context crafting — 10–15%
- Improve developer productivity with GitHub Copilot — 10–15%
- Configure privacy, content exclusions and safeguards — 10–15%
The study guide says most questions cover generally available (GA) features, but commonly used preview features can appear. This course follows these six domains; the large “features” domain is split across modules 2, 3, 4 and 8.
2. Copilot in the IDE: suggestions and Chat
To use Copilot in an editor you need three things: a GitHub account with Copilot access (a personal plan or a seat from an organization), the Copilot extension or plugin for your IDE, and a signed-in session. Supported editors include VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Vim/Neovim and others; feature coverage differs by editor, so chat and agent features may be missing in some.
- Install — add the GitHub Copilot extension from the editor’s marketplace.
- Sign in — authorize the editor with the GitHub account that holds the seat.
- Check status — the Copilot icon in the status bar shows whether it is active, disabled for the language, or blocked.
- Code — suggestions start as you type.
Inline suggestions appear as grey “ghost text”. In VS Code you press Tab to accept, Esc to dismiss, Alt+] / Alt+[ (Option on macOS) to cycle alternatives, and Ctrl+→ to accept word by word. A descriptive comment or function signature is a strong trigger. Next edit suggestions go further: after you change one line, Copilot predicts the related edit elsewhere in the file and lets you jump to it with Tab.
3. Copilot CLI, agent mode, MCP and the coding agent
GitHub Copilot CLI brings Copilot’s agent to the terminal. It can answer questions, write and debug code, generate scripts, manage files and work with GitHub.com (for example opening a pull request). It runs on Linux, macOS and Windows (PowerShell or WSL) and needs an active Copilot subscription; an organization or enterprise can switch it off by policy.
Install and sign in
- npm (Node.js 22 or later): npm install -g @github/copilot
- WinGet: winget install GitHub.Copilot · Homebrew: brew install --cask copilot-cli · or the install script / direct download
- Start it with copilot; on first launch use /login, or set a fine-grained token with the “Copilot Requests” permission in COPILOT_GITHUB_TOKEN, GH_TOKEN or GITHUB_TOKEN.
| Use | How |
|---|---|
| Interactive session | copilot, then chat; Shift+Tab switches to plan mode |
| One-shot / scripts | copilot -p "…" runs one task and exits; output can be piped in |
| Session commands | /model, /context, /compact, /mcp, /permissions, /feedback |
| Tool control | --allow-tool, --deny-tool (deny wins), --allow-all-tools |
By default the CLI asks before each tool use (shell command, file write, MCP tool). You can approve once, approve for the session, or reject with feedback. Session approvals do not carry over to new or resumed sessions. Launch it only in a trusted directory, never your home folder, because the agent can read and change files there.
4. Copilot on GitHub.com: reviews, Spaces, Spark and custom instructions
Copilot code review gives fast feedback on a pull request. You request it like any reviewer (choose Copilot in the Reviewers menu), or an owner configures automatic review, for example through a repository ruleset or a personal setting. Copilot reads the diff and leaves comments, often with suggested changes you can apply in one click.
| Copilot review does | Copilot review does not |
|---|---|
| Leave a “Comment” review with line comments | Approve the PR or block it with “Request changes” |
| Suggest fixes you can commit | Count toward required approvals |
| Follow your repository’s review instructions | Replace human or security review |
You can also select code in the IDE and ask Copilot to review it before you even push.
Customizable review standards
Teams encode their standards in instructions files: a repository-wide .github/copilot-instructions.md, and path-specific .github/instructions/NAME.instructions.md files whose front matter (applyTo) targets file patterns such as **/*.ts. Example rules: “flag any SQL built by string concatenation”, “public functions need docstrings”. Keep them short and concrete.
Pull request summaries
In the pull request description, the Copilot button generates a summary of the changes: an overview and a list of key changes per file. The author must still read and edit it; a summary can miss intent or overstate what the code does.
5. Copilot data flow and architecture
The exam asks you to “visualize the code suggestion lifecycle”. The official GitHub Copilot Fundamentals path describes it in stages:
- Trigger — you type, pause, write a comment or send a chat message.
- Gather context — the client collects what is relevant: code before and after the cursor, the file name and language, other open tabs (“neighbouring tabs”), imports, and for chat the selection, referenced files, chat history and any instructions files.
- Build the prompt — the context is ranked and trimmed to fit the model’s context window, then combined with the request.
- Send securely — the prompt travels over an encrypted connection to the Copilot service (a proxy in front of the models).
- Proxy filtering — the service checks the input, for example for toxic or off-topic content and attempts to manipulate the system prompt.
- Model inference — the large language model generates candidate completions or a chat answer.
- Post-processing — outputs are filtered again: harmful content, obvious quality problems, common insecure patterns such as hard-coded credentials, and code matching public code when your policy blocks it.
- Display and decision — you see ghost text or a chat reply and accept, edit or reject it.
6. Prompt engineering and context crafting
A prompt is everything Copilot receives: your words plus the context it gathers. A good prompt has four parts: the goal (what you want), the context (where and with what), the constraints (language, library, style, limits) and the expected output (format, example).
The Microsoft Learn Copilot path summarises good practice as the 4 S’s:
| Principle | Meaning | Example |
|---|---|---|
| Single | One clear task per prompt | “Write the validation function” before “write the tests” |
| Specific | Explicit, detailed instructions | “Validate a UK postcode; return true/false; no regex libraries” |
| Short | Concise while still specific | No storytelling, no repeated requirements |
| Surround | Give useful surrounding context | Descriptive file names, open related files, comments above the code |
How Copilot determines context
- In the editor: the current file around the cursor, other open tabs, file path and language.
- In chat: the selection, the active file, anything you add with #file, #selection or #codebase, participants like @workspace, the thread history and instructions files.
- On GitHub.com: the repository, issue or pull request you are viewing, and Spaces you attach.
7. Developer productivity, testing and security
The productivity domain is about using Copilot for the right jobs and keeping quality up. The study guide lists these use cases:
| Task | Typical Copilot move | Your check |
|---|---|---|
| Code generation | Comment or chat prompt, then inline completion | Tests pass, matches project style |
| Refactoring | Select code, ask to extract a function, rename, simplify | Behaviour unchanged; tests before and after |
| Documentation | /doc for docstrings; chat for README sections | Docs describe what the code really does |
| Learning | /explain on unfamiliar code, regex or SQL | Cross-check with official docs |
| Sample data | Generate realistic fixtures, seed data, JSON payloads | No real personal data; edge values included |
| Legacy modernization | Explain old code, then translate (e.g. COBOL or VB6 to Java or C#) | Behaviour-preserving tests, expert review |
Reducing context switching is a key benefit: instead of leaving the editor to search the web, open documentation or ask a colleague, you ask in chat or the terminal and stay in flow. That only helps if answers are checked; a wrong answer accepted quickly costs more than a slow search.
8. Administration, privacy and safeguards
With Copilot Business or Copilot Enterprise, owners manage Copilot centrally. Individual plans (Free, Pro, Pro+ and others) are managed by the user. Plans, prices and allowances change often, so check GitHub’s plans page rather than memorising numbers.
Policies
Organization owners decide which features members can use: Copilot in IDEs and on GitHub.com, Copilot Chat, Copilot CLI, the coding agent, Copilot code review, MCP servers, available models, and whether suggestions matching public code are allowed or blocked. In an enterprise, an enterprise policy set to enabled or disabled is enforced on every organization; when the enterprise leaves the choice open, each organization decides.
Seats
Owners grant access to all members or to selected teams and users. Removing a seat sets it to pending cancellation; access ends at the end of the billing cycle.
Audit log
- Copilot events cover plan, settings and policy changes, seat assignment (for example copilot.cfb_seat_assignment_created) and agent activity on GitHub.com.
- Filter with action:copilot; view agent activity with actor:Copilot.
- Events are kept for 180 days; stream to a SIEM for longer retention.
- Prompts that users send locally are not in the audit log.
The full lesson continues in the interactive course on this page.
Sources and references
- GitHub Copilot certification (Exam GH-300)
- Study guide for Exam GH-300: GitHub Copilot
- About GitHub Copilot CLI
- Content exclusion for GitHub Copilot
- GitHub Copilot code referencing
- Reviewing audit logs for GitHub Copilot
- REST API endpoints for Copilot user management
- Prompt engineering for GitHub Copilot
Unofficial practice course. Not affiliated with, endorsed by or sponsored by GitHub or Microsoft. GitHub Copilot and GH-300 are trademarks of their owners. Copilot features and plans change often: check the current exam study guide on Microsoft Learn before you book.
All courses · AI & Data - Free courses, certificate $5 | AMAADOR ACADEMY