12 GH-300 Practice Questions, One Scenario Per Domain

Instead of shuffled independent questions, each of GH-300's six domains gets one realistic scenario below, with two questions posed directly against it — closer to how the real exam layers multiple questions on a single situation.

16 min readUpdated September 2026GH-300 GitHub Copilot

GH-300 (GitHub Copilot) runs 100 minutes and is scheduled through Pearson VUE, testing six weighted domains from responsible AI use through organization-wide privacy configuration. Every question below is written the way GitHub actually writes them: a named person in a specific situation, followed by a decision they need to make — not a trivia definition to recall.

Click any question to reveal the answer choices, the correct one, and why each wrong option is wrong. Two questions share every scenario below, mirroring how the real exam can ask more than one thing about the same setup.

Want the full domain-by-domain breakdown first? See the GH-300 study guide.

Missing both questions in one domain below is a clearer signal than missing one question in each of six — check the full GH-300 domain list to see exactly what that domain covers before your next study session.

Use GitHub Copilot Responsibly

15–20%

Questions 12 · Priya, Staff Engineer

Priya is asked to write the internal guidelines before her team rolls out GitHub Copilot company-wide at a healthcare software company. Legal wants to know how the tool aligns with the company's AI governance policy before signing off.
1Which principle should Priya cite when legal asks who is responsible for monitoring Copilot's suggestions after rollout and addressing problems that surface later?Tap to see the four answer choices and the correct answer →
A)Transparency
B)Accountability✓
C)Inclusiveness
D)Fairness

Accountability requires that the people who create and operate an AI system remain responsible for its behavior and continuously monitor performance — exactly what legal is asking who owns.

A: Transparency is about the system being understandable and auditable to its users, not about who owns ongoing monitoring.

C: Inclusiveness concerns broadening who benefits from and can use a system, unrelated to ownership of outcomes.

D: Fairness concerns equal treatment across groups, not monitoring responsibility.

Key concept: Accountability = a person or team owns outcomes and monitors over time. Transparency = the system itself is legible to the people using it. These two get swapped constantly under time pressure.

2A junior developer on Priya's team assumes that because Copilot generated a code suggestion, it must be safe to merge without review. Which requirement does this violate?Tap to see the four answer choices and the correct answer →
A)The requirement to validate AI output before relying on it✓
B)The fairness principle
C)The requirement to configure content exclusions
D)The inclusiveness principle

The "validate and operate AI tools" subskill explicitly requires explaining the need to validate AI output — Copilot's suggestions aren't guaranteed correct or safe, and skipping review contradicts this requirement directly.

B: Fairness is about equal treatment across groups, unrelated to code-review discipline.

C: Content exclusions control what Copilot can see as context, not whether output needs review once generated.

D: Inclusiveness is about broadening access and benefit, unrelated to review requirements.

Key concept: "Copilot suggested it" is never sufficient justification on this exam — validation is a named, tested requirement, not just good practice.

Use GitHub Copilot Features

25–30%

Questions 34 · Marcus, Platform Team Lead

Marcus's team has used only inline Copilot suggestions for a year. He wants to modernize how they work with Copilot to handle larger, multi-file changes and routine issue triage.
3Marcus wants Copilot to autonomously plan and carry out a multi-step refactor across several files, deciding on its own which files need changes and running any necessary commands. Which capability should he use?Tap to see the four answer choices and the correct answer →
A)Copilot inline completions
B)Agent Mode✓
C)Copilot Edits
D)Copilot CLI

Agent Mode is built for exactly this — autonomous multi-step planning and execution across files with minimal step-by-step direction from the developer.

A: Inline completions only suggest the next few lines as you type — no autonomous multi-file planning involved.

C: Copilot Edits requires you to specify which files are in scope up front, which is more directed than what Marcus described.

D: The CLI operates from the terminal for scripts and file management, not multi-file autonomous refactors inside the IDE.

Key concept: Autonomous, multi-step, and Copilot decides the scope itself = Agent Mode. Directed, and you choose the files = Copilot Edits.

4Marcus also wants routine bug-fix issues picked up, worked, and turned into a pull request without a developer driving each step. Which feature fits?Tap to see the four answer choices and the correct answer →
A)Coding Agent✓
B)Copilot Chat
C)Sub-Agents
D)GitHub Copilot CLI

Coding Agent is the assigned-issue workflow — Copilot works a GitHub issue end-to-end and opens a pull request for review, matching exactly what Marcus wants for routine bug-fix triage.

B: Copilot Chat is a conversational interface for questions and explanations, not an autonomous issue-to-PR workflow.

C: Sub-Agents delegate part of an existing task to keep the main session's context focused — they aren't themselves an issue-to-PR pipeline.

D: The CLI is a terminal client for scripts and files, unrelated to issue-driven PR creation.

Key concept: When a scenario says "assign it a GitHub issue and it opens a PR," that's Coding Agent — not Agent Mode, which runs inside your active editing session.

Understand GitHub Copilot Data and Architecture

10–15%

Questions 56 · Aisha, Application Security Engineer

Aisha is reviewing whether GitHub Copilot is safe to approve for use on a proprietary codebase. She wants to understand what actually happens between typing code and seeing a suggestion.
5Aisha wants to know which stage happens immediately before a generated suggestion is displayed in the editor. Which of the following is that stage?Tap to see the four answer choices and the correct answer →
A)Input processing
B)Prompt building
C)Proxy filtering and post-processing✓
D)Context determination

After the model generates a candidate suggestion, it passes through proxy filtering and post-processing — including checks like public-code matching — before it's displayed. This is the last stage before the developer sees anything.

A: Input processing happens early, turning your code and context into usable input for the pipeline.

B: Prompt building happens next, assembling that processed input into the actual prompt sent to the model.

D: Context determination decides what surrounding code and history to include, and happens before the prompt is built.

Key concept: Order to memorize: context determined → input processed → prompt built → model generates → proxy filtering/post-processing → suggestion shown.

6Aisha is concerned Copilot might confidently suggest code that looks correct but subtly doesn't work. Which concept does this describe?Tap to see the four answer choices and the correct answer →
A)A content exclusion failure
B)A limitation inherent to LLMs✓
C)A responsible AI violation
D)A prompt engineering error

The "lifecycle and limitations" subskill explicitly covers limitations inherent to LLMs and to Copilot — including confident-but-wrong output, a well-documented property of generative models, not a misconfiguration.

A: Content exclusions control what Copilot can see as context — unrelated to output confidence or correctness.

C: This is a technical limitation, not itself a violation of a responsible-AI principle, though it's exactly why validation matters.

D: A prompt engineering error would be about how a request was phrased, not an inherent limitation of the underlying model.

Key concept: "Confidently wrong" output is a named LLM limitation on this exam. The correct response is always validation, not assuming something was misconfigured.

Apply Prompt Engineering and Context Crafting

10–15%

Questions 78 · Tomas, Junior Developer

Tomas keeps getting inconsistent results from Copilot Chat when he asks it to "write a function to process the data." A senior teammate suggests he change how he prompts.
7What is the senior teammate most likely recommending Tomas add to get more consistent results?Tap to see the four answer choices and the correct answer →
A)A few-shot example, like a sample input and expected output✓
B)A longer variable name
C)A zero-shot instruction with more exclamation points
D)Removing all context from the request

Few-shot prompting — supplying a concrete example such as sample input/output or a reference test — is GitHub's own recommended way to demonstrate the desired output shape when a bare instruction is too vague to be consistent.

B: Variable naming has no bearing on prompt consistency.

C: Zero-shot means no example at all, the opposite of what would help — and punctuation has no effect on prompt quality.

D: Removing context makes vague prompts worse, not better.

Key concept: Zero-shot = instruction only. Few-shot = instruction plus example(s). Vague zero-shot prompts are the classic "inconsistent results" root cause on this exam.

8Tomas rewrites his prompt as just: "Process the data." What best describes this prompt?Tap to see the four answer choices and the correct answer →
A)Few-shot prompting
B)Zero-shot prompting✓
C)Context crafting
D)A prompt engineering best practice

A bare instruction with no examples and no additional structure is zero-shot prompting by definition.

A: Few-shot requires at least one example — none was given here.

C: Context crafting means deliberately supplying relevant surrounding information, which this prompt does not do.

D: This prompt is the opposite of best practice — GitHub's guidance favors general-to-specific structure and examples, neither of which are present.

Key concept: A single vague sentence with nothing else is a textbook case of zero-shot prompting done poorly, not a trick term to overthink.

Improve Developer Productivity with GitHub Copilot

10–15%

Questions 910 · Elena, Team Lead

Elena's team uses Copilot to generate unit tests for a legacy billing module before refactoring it, to reduce the risk of breaking existing behavior.
9After Copilot generates a full suite of unit tests for the legacy module, what should Elena's team do before relying on them as a safety net for the refactor?Tap to see the four answer choices and the correct answer →
A)Merge them immediately since Copilot generated them from the existing code
B)Review the generated tests for coverage gaps and incorrect assumptions before trusting them✓
C)Delete them and write tests manually instead
D)Disable Copilot for the rest of the refactor

Generated tests, like any Copilot output, need human review — the productivity domain explicitly pairs test generation with identifying edge cases and validating assertions, not blind trust in generated coverage.

A: Skipping review is exactly the failure mode this exam tests against — "Copilot generated it" is never sufficient justification on its own.

C: Discarding useful generated work and starting from scratch ignores the actual value Copilot adds; the correct move is review, not replacement.

D: Disabling Copilot doesn't address the actual need, which is verifying the tests it already produced.

Key concept: This question tests Domain 5 material through a Domain 1 lens — "generate, then a human verifies" is the pattern the exam consistently rewards.

10Elena also asks Copilot to suggest performance improvements while modernizing the billing module. Which productivity capability is she using?Tap to see the four answer choices and the correct answer →
A)Security and performance suggestions✓
B)Public code filtering
C)Content exclusion configuration
D)Sub-Agent delegation

"Suggest security improvements and performance optimizations" is explicitly listed under the productivity domain's testing-and-security subskill.

B: Public code filtering is a privacy and safeguards policy, unrelated to generating performance suggestions.

C: Content exclusion controls what Copilot can see, not what kind of suggestions it generates.

D: Sub-Agent delegation is an organizational technique for managing context in a larger task, not a category of suggestion.

Key concept: When a scenario says "suggest improvements to speed, efficiency, or security," that's this named productivity subskill — not an architecture or privacy topic.

Configure Privacy, Content Exclusions, and Safeguards

10–15%

Questions 1112 · Devon, Platform Administrator

Devon is rolling out Copilot org-wide. One repository contains hardcoded credentials in a config folder that hasn't been cleaned up yet, and leadership wants to know whether Copilot's suggestions across the org will end up resembling public open-source code.
11Devon wants to stop Copilot from using the credentials-containing config folder as context anywhere it's referenced. What should Devon configure?Tap to see the four answer choices and the correct answer →
A)A content exclusion for that folder✓
B)A public-code-filtering policy
C)An audit log alert
D)A responsible AI principle

Content exclusions are exactly the mechanism for stopping Copilot from using specified files or paths as context, configurable at the repository or organization level.

B: Public-code filtering controls whether suggestions matching public code are blocked — it doesn't stop Copilot from reading a specific folder.

C: Audit log events record configuration changes for governance review; they don't prevent Copilot from accessing content.

D: Responsible AI principles are conceptual guidance, not a configuration mechanism.

Key concept: "Stop Copilot from seeing or using X" = content exclusion. "Stop suggestions that resemble public code" = public-code filtering. Two different problems, two different settings.

12Leadership specifically wants to know if organization-wide public-code filtering is guaranteed to apply the same way to every team. Is it?Tap to see the four answer choices and the correct answer →
A)Yes, it is a fixed platform-wide default that cannot be changed
B)No — an organization sets its own policy, but an enterprise administrator can lock a policy that overrides an organization's choice✓
C)No, only individual developers can set this on their own accounts
D)Yes, but only for GitHub Enterprise Cloud customers

Each organization can choose to allow or block suggestions matching public code, but when an enterprise administrator has explicitly configured a policy at the enterprise level, that policy overrides what an individual organization would otherwise set.

A: It's configurable, not a fixed, unchangeable default.

C: Policy control sits at the organization and enterprise level, not with individual developer accounts.

D: The enterprise-override behavior is a GitHub Enterprise Cloud capability, but the key fact tested is the override hierarchy itself, not a plan restriction.

Key concept: This is the same enterprise-overrides-organization hierarchy that shows up across multiple domains on this exam — memorize it once, apply it everywhere.

How to use this: the real GH-300 exam runs roughly 60 scored questions across all six domains. If you missed both questions in any domain above, spend another pass through that domain in the study guide before scheduling — the exam leans hard on applying a concept to a new situation, not recalling a definition.

Priya, Marcus, and the Rest of the Bank

MSCertQuiz has 500 GH-300 questions covering Agent Mode vs. Copilot Edits vs. CLI decisions, prompt engineering scenarios, data-architecture ordering, and org-policy hierarchies — with the same reveal-as-you-go format.

MSCertQuiz sells practice-exam access for GH-300 and other Microsoft and GitHub certifications; these questions are written by the same team that builds those question banks, checked against Microsoft's official GH-300 study guide.