GitHub Copilot Certification (GH-300): Exam Guide and 100 Practice Questions

The GitHub Copilot certification (exam GH-300) checks that you can use Copilot well and safely: responsible AI, agent mode, Copilot CLI, code review, prompt engineering, how Copilot handles data, and the settings that keep sensitive code out of it. It's an intermediate-level exam for developers, DevOps engineers, and technology managers who already know GitHub basics.

This guide covers the exam as of its August 7, 2026 skills update, then gives you 100 original practice questions weighted like the real exam. Each one hides its answer and an explanation until you open it. Copilot changes quickly, so every product detail here was checked against GitHub's documentation in October 2026. These questions are written for practice; they aren't taken from the real exam.

The exam at a glance

Exam code: GH-300, for the GitHub Copilot certification. Level: intermediate.
Time: 100 minutes. If the exam isn't offered in your language, you can request 30 extra minutes.
Passing score: 700 or higher.
Delivery: proctored, scheduled through Pearson VUE from the Microsoft Learn certification page. Price depends on the country where you take it.
Languages: English, Spanish, Portuguese (Brazil), Korean, and Japanese.
Retakes: you can retake 24 hours after a first failed attempt; later retakes have longer waits.
Validity: 2 years. GitHub is moving its certifications to Microsoft's recertification process, and certifications that expire before that's in place are being extended by 6 months.
Free extras: an official practice assessment and an exam sandbox that shows the real question interface.

Unlike GitHub Foundations, which doesn't expire, GH-300 needs renewing. Check the official pages linked at the end before you book, since details change.

The domains and their weights

Horizontal bar chart of GH-300 domain weights: responsible use 15 to 20 percent; Copilot features 25 to 30 percent; data and architecture, prompt engineering, developer productivity, and privacy and safeguards 10 to 15 percent each — Copilot features carries the most weight by a clear margin. The table below has the same numbers.

Domain	Exam weight	Practice questions here
Use GitHub Copilot responsibly	15–20%	18
Use GitHub Copilot features	25–30%	28
Understand GitHub Copilot data and architecture	10–15%	13
Apply prompt engineering and context crafting	10–15%	13
Improve developer productivity with GitHub Copilot	10–15%	14
Configure privacy, content exclusions, and safeguards	10–15%	14

One oddity: Microsoft's "skills at a glance" list shows the features domain twice, as "Use GitHub Copilot features" and "GitHub Copilot features", both at 25–30%. Only one of them has any skills listed, so it reads as a duplicate, and this guide treats the exam as six domains. The August 2026 update made minor changes to the IDE, features, and safeguards objectives. Newer topics include Copilot CLI, agent sessions and sub-agents, Copilot Spaces, Spark, instructions files, and managing seats through the REST API.

How to use these questions

Answer each question before opening its answer, and tally your misses per domain. If you get more than about 80% right, take the official practice assessment next. Several questions describe features that change often; when one surprises you, check GitHub's docs instead of relying on memory, because that's what the exam rewards.

Use GitHub Copilot responsibly (15–20%)

Questions 1–18.

Why do you need to review code that Copilot generates before using it?
1. Copilot output is always identical to its training data
2. Reviewing improves the model's accuracy for everyone
3. Large language models can produce code that looks plausible but is incorrect, insecure, or outdated
4. GitHub requires a manual approval for each suggestion
Show answer

Answer: C. Copilot predicts likely code; it doesn't verify it. GitHub's own docs warn that suggestions may look valid but be semantically or syntactically wrong.
What does it mean when an AI tool "hallucinates"?
1. It refuses to answer
2. It produces confident output that is false, such as a function or API that doesn't exist
3. It repeats your prompt back
4. It runs slowly
Show answer

Answer: B. Hallucinations are why you check unfamiliar APIs against real documentation before trusting them.
Copilot suggests a database query built by concatenating user input into a SQL string. What's the best response?
1. Accept it; Copilot filters insecure code, so it must be safe
2. Disable Copilot for the file
3. Accept it and add a code comment
4. Reject or rewrite it with parameterized queries, then test it
Show answer

Answer: D. Content filters reduce insecure suggestions but don't guarantee secure code. You're responsible for spotting injection risks.
Copilot's training data comes from many sources. What risk does that create?
1. Suggestions are always in English
2. Suggestions only work for JavaScript
3. Biases and errors in the training data can show up in suggestions
4. Copilot can only suggest code written before 2010
Show answer

Answer: C. GitHub's responsible-use documentation lists bias carried over from training data as a known limitation.
A developer commits code that Copilot wrote and it causes a production outage. Who is accountable?
1. The developer and team who reviewed and shipped the code
2. GitHub
3. No one, because AI wrote it
4. The model provider
Show answer

Answer: A. Accountability stays with people. Copilot is a tool; the person who accepts, reviews, and merges the code owns it.
Which of these is not one of Microsoft's six responsible AI principles?
1. Accountability
2. Fairness
3. Transparency
4. Profitability
Show answer

Answer: D. The six principles are fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
In responsible AI, what does the transparency principle ask for?
1. Open-sourcing every model
2. Helping people understand how an AI system works, what it's for, and its limitations
3. Making AI output free
4. Publishing all user prompts
Show answer

Answer: B. Users should know when they're working with AI output and what it can and can't do.
A new team member accepts Copilot suggestions they don't understand. What's the main risk?
1. Bugs and security issues slip through, and they don't build the understanding needed to maintain the code
2. Copilot will stop working for them
3. Their suggestions will be blocked
4. Their code will run slower
Show answer

Answer: A. Over-reliance is a recognized harm. Ask Copilot Chat to explain the code, and read it before accepting.
What should you avoid pasting into a Copilot Chat prompt?
1. Test output
2. Error messages
3. Secrets such as API keys, passwords, and tokens
4. Function names
Show answer

Answer: C. Treat prompts like any data you'd send to a service: never include credentials. Use placeholders instead.
Copilot suggests an API call that was deprecated in the latest version of a library. Why might this happen?
1. Deprecated APIs are preferred for safety
2. Models learn from code written over time and may not reflect the newest library versions
3. Copilot reads your package registry live
4. Your editor is out of date
Show answer

Answer: B. Check suggestions against current documentation, especially for fast-moving libraries.
Copilot cloud agent has opened a pull request with a feature you asked for. What's the responsible next step?
1. Approve it without reading, since you wrote the prompt
2. Merge it immediately; agents test their own work
3. Close it and write the code yourself
4. Review it like any pull request: read the diff, run the checks, and request changes if needed
Show answer

Answer: D. Human review stays in the loop. Agent pull requests go through the same review and checks as a teammate's.
In agent mode, Copilot proposes running a terminal command that deletes a directory. What should you do?
1. Restart the IDE
2. Disable agent mode permanently
3. Read and understand the command before approving it
4. Approve all commands automatically to save time
Show answer

Answer: C. Agent mode can run commands, so treat each one as if you were typing it yourself.
Copilot generates unit tests that all pass on the first run. What should you check?
1. That the tests actually assert meaningful behavior and cover edge cases, not just that they run
2. Nothing; passing tests prove the code is correct
3. That the test file name is long enough
4. That Copilot used the newest model
Show answer

Answer: A. Generated tests can be shallow or assert whatever the current code does, bugs included.
Which mitigation best reduces the risk of unknowingly using code that matches licensed public code?
1. Turn off syntax highlighting
2. Only use Copilot on weekends
3. Rename variables after accepting suggestions
4. Block suggestions matching public code, or use code referencing to see matches and their licenses
Show answer

Answer: D. Depending on your policy, matches are blocked or shown with information about the matching code and its license.
Copilot writes a form validation function that rejects names containing apostrophes or accents. Which responsible AI principle is most at stake?
1. None; it's a style choice
2. Inclusiveness and fairness
3. Transparency
4. Reliability only
Show answer

Answer: B. Assumptions baked into generated code can exclude real users. Reviewing for them is part of using AI responsibly.
Which statement about Copilot's limitations is correct?
1. It has limited context and can miss project-specific requirements unless you provide them
2. It understands your entire codebase perfectly
3. It guarantees code is free of public-code matches
4. It never makes syntax errors
Show answer

Answer: A. Copilot only knows what's in its context. Instructions files, open files, and specific prompts close the gap.
A teammate argues that because Copilot passed a security review once, its suggestions no longer need review. What's wrong with this?
1. Security reviews expire after a week
2. Copilot only reviews its own code
3. Every suggestion is newly generated and can differ, so each one needs review
4. Nothing; one review is enough
Show answer

Answer: C. Model output varies from request to request. Review is per change, not per tool.
What's the best description of how to operate Copilot responsibly day to day?
1. Accept most suggestions to keep the model warm
2. Treat it as an assistant: give clear context, review everything, test it, and keep human judgment in charge
3. Use it only for comments
4. Let it merge its own pull requests
Show answer

Answer: B. Copilot speeds you up, but you still make the decisions.

Use GitHub Copilot features (25–30%)

Questions 19–46.

Diagram of a repository tree showing Copilot customization files: .github/copilot-instructions.md for repository-wide instructions, .github/instructions/NAME.instructions.md for path-specific instructions with applyTo, .github/prompts/NAME.prompt.md for reusable prompts run with a slash command, and AGENTS.md files where the nearest one in the tree wins — Four customization files, four jobs. Several questions in this section test which one to use.

What do you need to start using Copilot in VS Code?
1. Nothing; it's built into every copy of VS Code without sign-in
2. A GitHub Enterprise Server instance
3. A self-hosted model server
4. The GitHub Copilot extension and a GitHub account with Copilot access
Show answer

Answer: D. Sign in with a GitHub account that has a Copilot plan. Copilot isn't available for GitHub Enterprise Server.
Copilot shows a grey inline suggestion as you type. Which key accepts it?
1. F5
2. Shift
3. Tab
4. Enter
Show answer

Answer: C. Tab accepts, Esc dismisses. You can also accept word by word.
In VS Code, how do you open Copilot chat directly inside the editor, at your cursor?
1. Ctrl+I (Cmd+I on macOS)
2. Double-click the file tab
3. Ctrl+Shift+P, then type "git"
4. Alt+F4
Show answer

Answer: A. Inline chat edits code in place. The Chat view is better for longer conversations.
What can Copilot agent mode do that ask mode can't?
1. Answer questions
2. Show syntax highlighting
3. Use a model at all
4. Plan and make changes across multiple files, run terminal commands, and iterate until the task is done
Show answer

Answer: D. Agent mode works autonomously on a task, with you approving commands and reviewing edits.
What is the Model Context Protocol (MCP) used for with Copilot?
1. Encrypting chat history
2. Connecting Copilot's agent to external tools and data sources, such as the GitHub MCP server or a database
3. Compressing prompts
4. Switching themes
Show answer

Answer: B. MCP is an open protocol. MCP servers give agent mode new tools it can call.
In VS Code, which file configures MCP servers for a specific workspace?
1. .vscode/mcp.json
2. package.json
3. .github/workflows/mcp.yml
4. .github/copilot-instructions.md
Show answer

Answer: A. Committing .vscode/mcp.json shares the same MCP setup with everyone who opens the workspace.
Why delegate part of a task to a sub-agent?
1. Sub-agents don't need a Copilot plan
2. Sub-agents skip code review
3. The sub-agent works in its own context, which keeps the main session's context focused and uncluttered
4. Sub-agents run on cheaper hardware
Show answer

Answer: C. The study guide pairs sub-agents with "optimized context usage": a side task doesn't fill the main conversation with detail.
How do you install GitHub Copilot CLI with npm?
1. gh install copilot
2. npm install -g @github/copilot
3. npm install copilot-chat
4. pip install github-copilot
Show answer

Answer: B. It needs Node.js 22 or later. Homebrew, WinGet, and an install script are alternatives.
After installing Copilot CLI, which command starts its interactive interface?
1. git copilot
2. code --copilot
3. gh run
4. copilot
Show answer

Answer: D. On first launch, if you aren't signed in, it asks you to use /login.
In Copilot CLI, what does the /delegate command do?
1. Deletes the current session
2. Assigns an issue to a teammate
3. Hands the work to Copilot in the cloud, which creates a pull request on the remote repository
4. Switches the AI model
Show answer

Answer: C. Use it when a task can finish in the background as a pull request instead of in your terminal.
Which Copilot CLI slash command lets you choose which AI model to use?
1. /model
2. /clear
3. /mcp
4. /resume
Show answer

Answer: A. /clear starts a new conversation, /mcp manages MCP servers, and /resume picks up a previous session.
Which Copilot CLI slash command manages MCP server configuration?
1. /login
2. /model
3. /help
4. /mcp
Show answer

Answer: D. /mcp can list, add, edit, enable, or disable MCP servers.
You want to return to a Copilot CLI conversation you had yesterday. What do you use?
1. /clear
2. The /resume command, or start with --continue
3. Reinstall the CLI
4. Copy the old terminal output back in
Show answer

Answer: B. Sessions are saved, so you can pick work back up with its context intact.
Copilot CLI won't start for a developer whose colleagues use it fine. What's a likely cause?
1. Their organization or enterprise has disabled Copilot CLI in its policies
2. They have too many repositories
3. CLI only works on Fridays
4. Their Node.js is too new
Show answer

Answer: A. Organization owners and enterprise administrators control CLI availability with policies.
How do you ask Copilot to review a pull request on GitHub.com?
1. Mention @copilot in a commit message
2. Add a review.yml workflow
3. Request Copilot as a reviewer on the pull request
4. Copilot reviews every PR automatically with no setup
Show answer

Answer: C. By default, Copilot only reviews a pull request when you request it. Automatic reviews can be turned on in settings or rulesets.
Does an approving review from Copilot count toward a branch's required approvals?
1. Only for admins
2. Not by default; it only counts if Copilot approvals are explicitly enabled
3. Only on public repositories
4. Yes, always
Show answer

Answer: B. By default, Copilot's reviews don't count toward required approvals.
How can an organization run Copilot code review automatically on every new pull request?
1. Pin Copilot to the repository
2. Add Copilot as an outside collaborator
3. Create a GitHub App
4. Enable automatic review in repository or organization settings, or with a ruleset
Show answer

Answer: D. Options include reviewing when a PR opens, on each new push, and on drafts.
Where do repository-wide custom instructions for Copilot live?
1. Repository Settings → General
2. .copilot/config.json
3. .github/copilot-instructions.md
4. README.md
Show answer

Answer: C. They're added to Copilot Chat requests and used by Copilot code review and the cloud agent.
You want one set of instructions to apply only to files under api/. What do you create?
1. A .github/instructions/api.instructions.md file with an applyTo glob in its frontmatter
2. A second copilot-instructions.md in api/
3. A branch protection rule
4. A CODEOWNERS entry
Show answer

Answer: A. Path-specific instruction files use the applyTo frontmatter key to target files.
A repository has an AGENTS.md at the root and another in src/. An agent edits a file in src/. Which file's instructions take precedence?
1. Neither; only one is allowed
2. The root one
3. Whichever was edited most recently
4. The one in src/, the nearest in the directory tree
Show answer

Answer: D. The nearest AGENTS.md wins, so subprojects can carry their own guidance.
What are prompt files for?
1. Storing chat history
2. Saving reusable prompts in .github/prompts/*.prompt.md, run in chat by typing / and the prompt name
3. Defining Actions workflows
4. Configuring content exclusions
Show answer

Answer: B. Prompt files give a team consistent, reusable prompts. Frontmatter can set the agent, model, and tools.
What are Copilot Spaces?
1. Collections of context (repositories, files, issues, pull requests, and notes) that ground Copilot's answers, and that you can share
2. Folders on a self-hosted runner
3. Codespaces with Copilot preinstalled
4. Private chat rooms for teams
Show answer

Answer: A. Spaces let teams curate the knowledge Copilot should draw on for a topic or project.
What is GitHub Spark?
1. A Copilot plan
2. A code search index
3. A tool for building and deploying small apps from natural-language descriptions
4. A CI runner
Show answer

Answer: C. You describe an app; Spark builds it, and you refine it with more prompts or code.
What can Copilot generate for you when you open a pull request on GitHub.com?
1. A new branch protection rule
2. A summary of the pull request's changes for the description
3. A license file
4. A release
Show answer

Answer: B. Pull request summaries save writing time, but check them for accuracy before posting.
In Copilot Chat in VS Code, which slash command generates tests for the selected code?
1. /clear
2. /new
3. /explain
4. /tests
Show answer

Answer: D. /explain explains code, /fix proposes fixes, and /tests writes tests.
A Copilot Chat answer was unhelpful. What's the built-in way to tell GitHub?
1. Open an issue on your repository
2. Uninstall the extension
3. Use the thumbs-down feedback on the response
4. Email the model provider
Show answer

Answer: C. Feedback controls on chat responses are the intended channel.
An organization owner wants to stop members from using Copilot in the CLI while still allowing it in IDEs. Where is that controlled?
1. The organization's Copilot policies
2. The repository's README
3. Each developer's IDE settings
4. A .gitignore rule
Show answer

Answer: A. Organization policies control feature availability across IDEs, the CLI, and GitHub.com.
Which REST API endpoint lists every Copilot seat assigned in an organization?
1. POST /orgs/{org}/copilot/billing/selected_users
2. GET /repos/{owner}/{repo}/copilot
3. GET /user/copilot
4. GET /orgs/{org}/copilot/billing/seats
Show answer

Answer: D. POST and DELETE on /orgs/{org}/copilot/billing/selected_users add and remove users.

Understand GitHub Copilot data and architecture (10–15%)

Questions 47–59.

Diagram of the life of an inline suggestion: gather context, build the prompt, the model generates, filters block harmful or insecure content and handle public-code matches, then the developer accepts or rejects; plus the public-code check details: about 150 characters, fewer than 1 percent of suggestions — Context, prompt, generation, filtering, then your decision. Questions in this section test the order and the numbers.

For an inline suggestion, which context does Copilot use?
1. Every repository in your organization
2. Code around your cursor, plus related context such as snippets from files open in other editor tabs
3. Your browser history
4. Only the current line
Show answer

Answer: B. That's why opening related files and closing unrelated ones improves suggestions.
What is the "prompt" in Copilot's architecture?
1. The context Copilot assembles (your code, instructions, and request) and sends to the model
2. The model's training data
3. Only the text you type in chat
4. A configuration file
Show answer

Answer: A. Even inline suggestions have a prompt, built automatically from your editor's context.
Which statement about how Copilot filters generated output is correct?
1. Filters run only on Copilot Free
2. There's no filtering; output goes straight to the editor
3. Content filters detect and block harmful or offensive content and insecure code, and public-code matches can be blocked or annotated
4. Only spelling is checked
Show answer

Answer: C. GitHub's responsible-use documentation describes these filters on generated suggestions.
When checking for matches with public code, how much surrounding code does Copilot compare along with the suggestion?
1. Exactly one line
2. About 150 characters
3. The whole file
4. The whole repository
Show answer

Answer: B. Copilot compares the suggestion plus roughly 150 characters of surrounding code against an index of public repositories on GitHub.
Roughly how often do Copilot suggestions match public code?
1. In about half of suggestions
2. Never
3. In every suggestion
4. In fewer than 1% of suggestions
Show answer

Answer: D. Matches are uncommon, which is why a filter or reference is enough to manage them.
With code referencing enabled, what does Copilot record when you accept a matching inline suggestion unchanged?
1. Nothing
2. Your IP address
3. The matching file URLs and any detected licenses
4. The author's email
Show answer

Answer: C. Only accepted, unchanged suggestions are checked this way. Edited suggestions aren't.
Which code is not part of the index Copilot checks for public-code matches?
1. Private repositories and code hosted outside GitHub
2. Code in public forks on GitHub
3. Public repositories on GitHub.com
4. Popular open source projects on GitHub
Show answer

Answer: A. The index covers public repositories on GitHub.com only.
Which order correctly describes the life of an inline suggestion?
1. Filters run → you accept → model generates → context gathered
2. You accept → context gathered → prompt built → model generates
3. Model generates → context gathered → you accept → filters run
4. Context gathered → prompt built → model generates → filters run → suggestion shown for you to accept or reject
Show answer

Answer: D. Context first, then generation, then filtering, then your decision.
Why can't Copilot simply read your entire large codebase for every suggestion?
1. It only reads READMEs
2. Models have a limited context window, so only the most relevant context fits
3. Large codebases are always excluded
4. It's not allowed to read code
Show answer

Answer: B. Context limits are why instructions files, Spaces, and specific prompts matter.
You send the same prompt to Copilot twice and get two different answers. Why?
1. Language model output isn't deterministic; generation can vary between requests
2. The second request used no context
3. One answer was cached
4. Your plan was downgraded
Show answer

Answer: A. Variation is normal. It's one more reason every result needs review.
Inline suggestions are designed to produce what kind of output?
1. Images
2. Long natural-language essays
3. Code and code-related output
4. Audio
Show answer

Answer: C. Inline suggestions target code. Use Copilot Chat for explanations in natural language.
In a Copilot Chat conversation, how do earlier messages affect a new question?
1. They're sent to other users
2. Earlier turns in the same conversation are used as context for the new question
3. They're ignored
4. Only the very first message is used
Show answer

Answer: B. Start a new conversation when you change topics, so old context doesn't steer the new answer.
A suggestion appears in a language you rarely use and looks lower quality. What explains this?
1. The editor theme affects quality
2. Copilot deliberately degrades rare languages
3. Copilot only supports Python
4. Suggestion quality depends on how much of that language was in the training data
Show answer

Answer: D. GitHub notes that the composition of training data affects quality by language.

Apply prompt engineering and context crafting (10–15%)

Questions 60–72.

What is zero-shot prompting?
1. Sending an empty prompt
2. Running the prompt zero times
3. Asking the model to do a task with no examples, only instructions
4. Giving the model several input/output examples first
Show answer

Answer: C. Zero-shot relies entirely on the instruction and context.
What is few-shot prompting?
1. Including a few examples of the desired input and output in the prompt so the model follows the pattern
2. Limiting the response to a few words
3. Using a small model
4. Asking several unrelated questions at once
Show answer

Answer: A. Examples are the fastest way to get output in a specific format or style.
Which prompt is most likely to get useful code from Copilot Chat?
1. "Write code"
2. "ZIP?"
3. "Do the thing from before"
4. "Write a TypeScript function that validates US ZIP codes, accepts 5-digit and ZIP+4 formats, and returns a boolean. Include JSDoc."
Show answer

Answer: D. Specific language, inputs, outputs, and constraints leave less for the model to guess.
A large feature request gives poor results in one prompt. What's the best technique?
1. Repeat the same prompt louder
2. Break it into smaller, sequential steps and prompt for each
3. Add more emoji
4. Switch to inline suggestions only
Show answer

Answer: B. Smaller tasks keep each request focused and easier to verify.
How can you steer inline suggestions without opening chat?
1. Write a clear comment describing the function before you start it, and use descriptive names
2. Turn off syntax highlighting
3. Rename the file randomly
4. Type faster
Show answer

Answer: A. Comments and names are part of the context Copilot reads.
Which action is most likely to improve inline suggestions for a module you're writing?
1. Restart your computer
2. Close all other files
3. Open the related files, such as the interface or types it uses, in other editor tabs
4. Clear the clipboard
Show answer

Answer: C. Snippets from open tabs are part of the context Copilot uses.
Why close unrelated files when working with Copilot?
1. Copilot charges per open tab
2. Unrelated open files can add noise to the context
3. Copilot only works with one tab
4. It frees GPU memory
Show answer

Answer: B. Relevant context helps; irrelevant context dilutes it.
Which elements make up a well-structured prompt?
1. A random seed
2. Only a code snippet
3. Only the desired output
4. The goal, relevant context, constraints, and, if needed, examples of the expected output
Show answer

Answer: D. Goal, context, constraints, and examples cover most of what the model needs.
In VS Code chat, how do you make sure a specific file is included as context?
1. Delete the other files
2. Move it to the repository root
3. Reference it explicitly, for example with #file, or attach it
4. Rename it to context.txt
Show answer

Answer: C. Explicit references beat hoping Copilot finds the right file on its own.
Copilot's first answer is close but uses the wrong library. What should you do?
1. Follow up in the same conversation and say which library to use
2. Report it as a bug
3. Start over with a completely different topic
4. Accept it and fix it later
Show answer

Answer: A. Iterating with follow-ups uses the conversation's history to refine the result.
You've finished one task in Copilot Chat and are starting an unrelated one. What's best practice?
1. Switch models every message
2. Keep going in the same conversation
3. Close VS Code
4. Start a new conversation so earlier context doesn't influence the new task
Show answer

Answer: D. Chat history is context. Old topics can steer new answers in the wrong direction.
You want Copilot to always use your team's naming conventions without repeating them in every prompt. What should you use?
1. Zero-shot prompting
2. A repository custom instructions file
3. A larger monitor
4. A longer chat history
Show answer

Answer: B. Instructions files add the same context to every request automatically.
Which is a prompt engineering principle?
1. Be specific and provide context; ambiguity produces guesses
2. Never give examples
3. Always ask several questions in one prompt
4. Vaguer prompts give more creative and therefore better code
Show answer

Answer: A. The clearer the request, the less the model has to infer.

Improve developer productivity with GitHub Copilot (10–15%)

Questions 73–86.

How can Copilot help most when you join a project with an unfamiliar codebase?
1. It deletes legacy code automatically
2. It can't help with existing code
3. Ask Copilot Chat to explain files, functions, and how pieces connect, without leaving the editor
4. It rewrites the whole project in your preferred language
Show answer

Answer: C. Explanations in the editor cut down on context switching to docs and search.
Which task is a good use of Copilot for documentation?
1. Inventing API behavior that doesn't exist
2. Generating docstrings and comments for existing functions, then reviewing them for accuracy
3. Writing the license
4. Removing all comments
Show answer

Answer: B. Copilot drafts quickly; you make sure it describes what the code actually does.
How can Copilot help when refactoring a long function?
1. Only rename variables
2. Merge the change automatically
3. It can't refactor
4. Suggest a cleaner structure, such as extracting helpers, while you verify behavior with tests
Show answer

Answer: D. Pair refactoring with tests so you know behavior didn't change.
You need 50 realistic but fake customer records for a demo. How can Copilot help?
1. Download data from the internet
2. Copy real customer data from production
3. Generate sample data in the format you need, such as JSON or CSV, with realistic-looking values
4. It can't generate data
Show answer

Answer: C. Synthetic sample data is a classic productivity win, and it keeps real personal data out of demos.
Your team must move an old application to a modern language and framework. How can Copilot help?
1. Explain the legacy code and propose modern equivalents step by step, with tests to confirm behavior
2. It only works on new projects
3. It guarantees a bug-free automatic conversion
4. It refuses legacy languages
Show answer

Answer: A. Modernization goes faster with Copilot, but each translated piece still needs verification.
How does Copilot help you learn a new programming language?
1. It can't
2. It replaces the need to learn
3. It only works in English
4. Explain idioms, translate snippets from a language you know, and answer follow-up questions in context
Show answer

Answer: D. Asking in the editor keeps you in flow instead of searching elsewhere.
Which request is best for getting unit tests from Copilot?
1. "Test"
2. Select a function and ask for unit tests that cover normal input, edge cases, and invalid input
3. Ask for tests without any code selected or referenced
4. Ask for tests for the entire repository at once
Show answer

Answer: B. Scope and explicit coverage goals produce focused, useful tests.
What's the difference between the unit and integration tests Copilot can generate?
1. Unit tests check one piece in isolation; integration tests check how components work together
2. There is none
3. Integration tests are only for front ends
4. Unit tests require a database
Show answer

Answer: A. You can ask Copilot for either. Be explicit about which one you want.
Which of these is an edge case Copilot might suggest testing for a function that divides two numbers?
1. Dividing 10 by 2
2. Dividing 100 by 10
3. Dividing by zero
4. Dividing 4 by 2
Show answer

Answer: C. Ask Copilot "what edge cases should this handle?" and then write assertions for them.
What is an assertion in a test?
1. A comment explaining the test
2. A check that the actual result matches the expected result
3. A random input
4. A test file name
Show answer

Answer: B. Tests without meaningful assertions pass without proving anything.
How can Copilot help improve the security of existing code?
1. It encrypts your repository
2. It can't comment on security
3. It disables insecure features automatically
4. Ask it to review code for vulnerabilities and suggest fixes, such as input validation or parameterized queries
Show answer

Answer: D. Combine Copilot's suggestions with code scanning and human review.
A function runs slowly on large input. How can Copilot help?
1. It can't analyze performance
2. It automatically adds more servers
3. Ask it to identify bottlenecks and suggest optimizations, such as a better algorithm, then measure the result
4. It deletes the function
Show answer

Answer: C. Always measure; a suggested optimization isn't proven until benchmarks agree.
How does Copilot reduce context switching for developers?
1. By answering questions and generating code inside the editor, so you leave it less to search docs or forums
2. By muting notifications
3. By closing your browser
4. By scheduling meetings
Show answer

Answer: A. Staying in the editor keeps you in flow.
Which statement best describes Copilot's effect on productivity?
1. It writes complete applications with no human input
2. It removes the need for code review
3. It only helps senior developers
4. It speeds up common tasks like boilerplate, tests, and docs, while developers still review and own the result
Show answer

Answer: D. The productivity gain is real; the responsibility doesn't move.

Configure privacy, content exclusions, and safeguards (10–15%)

Questions 87–100.

Who can configure content exclusions for Copilot?
1. Only GitHub Support
2. Repository administrators, organization owners, and enterprise owners
3. Only the person who installed Copilot
4. Any user with read access
Show answer

Answer: B. Repository admins exclude content in their repositories; organization owners set exclusions that apply to members with a Copilot seat.
What happens to a file that's excluded from Copilot?
1. Copilot won't use its content to generate suggestions or answers, and inline suggestions aren't offered in it
2. It's made private
3. It's encrypted
4. It's deleted
Show answer

Answer: A. Exclusions keep sensitive files, such as secrets and configuration, out of Copilot's context.
Where is content exclusion not supported?
1. Copilot CLI chat
2. Inline suggestions in VS Code
3. Edit and agent modes of Copilot Chat in VS Code and other editors
4. Copilot code review on GitHub.com
Show answer

Answer: C. This is an important limitation. Exclusions don't stop agent-mode edits, so sensitive files need other protections.
Which statement about content exclusion limitations is correct?
1. It applies to symbolic links
2. Copilot may still use semantic information about an excluded file that the IDE provides, such as types or hover definitions
3. It applies to repositories on remote filesystems
4. It encrypts the file contents
Show answer

Answer: B. Exclusions also don't apply to symbolic links or to repositories on remote filesystems.
Your organization wants to prevent Copilot from suggesting code that matches public code. What should you configure?
1. A branch protection rule
2. Two-factor authentication
3. A .gitignore entry
4. Block suggestions matching public code, in the Copilot policies
Show answer

Answer: D. Policies can block matching suggestions, or allow them with references.
What's the difference between blocking and allowing suggestions that match public code?
1. Blocking deletes your code
2. Allowing turns off all filters
3. Blocking hides matching suggestions; allowing shows them along with information about the matching code
4. No difference
Show answer

Answer: C. Allowing with references lets you decide, with license information to hand.
Who owns the code you write with Copilot's suggestions?
1. GitHub doesn't claim ownership of Copilot's suggestions; you're responsible for the code you accept and use
2. The model provider owns it
3. GitHub owns it
4. It's automatically public domain
Show answer

Answer: A. Check licenses, review, and test as you would for any code you're responsible for.
A developer stops getting any Copilot suggestions in one specific file. What's a likely cause?
1. Their keyboard is broken
2. Copilot only works on Mondays
3. The file is too short
4. The file matches a content exclusion
Show answer

Answer: D. Check content exclusions before deeper troubleshooting.
Copilot works in Python files but not in Markdown files for one user. What's a likely cause?
1. Their license expired for Markdown only
2. Copilot is disabled for that language in the editor settings
3. Content exclusion only affects Python
4. Markdown can't be edited in VS Code
Show answer

Answer: B. Editor settings can enable or disable Copilot per language.
Copilot suggestions stopped completely for a whole team after a network change. What should they check first?
1. Whether a proxy or firewall is blocking Copilot's network endpoints, and whether they're still signed in
2. Their monitor resolution
3. Their repository's license
4. Their Git username
Show answer

Answer: A. Copilot needs network access to GitHub's services. Corporate proxies are a common cause.
A repository admin adds a content exclusion. Whom does it affect?
1. Only users of Copilot Free
2. Only the admin
3. Copilot users working in that repository
4. Every GitHub user
Show answer

Answer: C. Repository-level exclusions apply to people using Copilot in that repository.
Does content exclusion also apply to Copilot code review on GitHub.com?
1. Only when you're signed out
2. Yes, Copilot code review respects content exclusions
3. Only for private repositories
4. No, never
Show answer

Answer: B. Excluded files are left out of Copilot's review.
Which editor setting can a developer change to reduce Copilot's influence while writing sensitive code?
1. Delete the .git folder
2. Uninstall Git
3. Change the theme
4. Disable Copilot for a language, or turn off inline suggestions temporarily
Show answer

Answer: D. Individual settings complement organization policies and exclusions.
Which approach best protects a file of production secrets in a repository where developers use Copilot?
1. Rely on Copilot's content filters
2. Make the repository public
3. Don't commit secrets; store them in a secrets manager or environment variables, and add a content exclusion for any sensitive config that must exist
4. Rename the file
Show answer

Answer: C. Exclusions limit what Copilot reads; keeping secrets out of the repository protects them everywhere.

Final tips for exam day

Features is a quarter of the exam or more. Know agent mode versus ask mode, what MCP is for, the Copilot CLI basics, and where each customization file lives.
Learn the exceptions. Content exclusion doesn't apply in Edit and agent modes. Copilot's code reviews don't count toward required approvals by default. These are exactly the details exam questions are built on.
Know the numbers: about 150 characters for the public-code check, fewer than 1% of suggestions matching, 700 to pass.
When responsible-AI questions offer "trust the tool" against "review and verify", choose review. The exam consistently keeps humans accountable.
Use it hands-on. Add a .github/copilot-instructions.md file, write a prompt file, try agent mode on a small task, and install Copilot CLI. An afternoon of practice covers a lot of the features domain.

Official resources

GitHub Copilot certification page (Microsoft Learn): scheduling, the free practice assessment, and exam policies.
Exam GH-300 study guide: the official skills list and change log.
GitHub Copilot Fundamentals, part 1 and part 2: free self-paced training.
GitHub Copilot documentation: the source of truth for every feature in these questions.

Preparing for more GitHub certifications? See GitHub Foundations (GH-900): Exam Guide and 100 Practice Questions, plus the hands-on Git and GitHub and GitHub Actions lesson posts.