Examworthyexamworthy.comGitHub Copilot cheat sheet
GitHub
Free to share. Examworthy is not affiliated with or endorsed by GitHub; GH-300 and related marks belong to their respective owners.
At a glance
Format: Multiple choice and multiple response
Domain weight map
Heaviest first - spend your time hereHow this exam thinks
GH-300 is a pick-the-right-surface exam: almost every question is a developer scenario with a stated need, and the right answer is the Copilot surface, plan, data-handling control, or prompting technique built for it - with the human always accountable for validating the output.
Spot the trap
Tempting wrong answers, and why they failTempting but wrong
Agent Mode is the right choice when you only want existing code explained in words.
Why it fails
Wrong. Agent Mode is built to take autonomous action and edit files, tracing call graphs and applying refactors it judges necessary. That exceeds and contradicts a read-only request. For a prose explanation with no edits, Copilot Chat is the fit.
Use GitHub Copilot features
Tempting but wrong
A plausibly named Copilot helper that may not exist is an example of bias in the output.
Why it fails
Bias concerns unfair treatment learned from skewed training data, where one group of inputs is handled less fairly than another. A fabricated helper that may not exist is about invented output, which is hallucination, not unfairness.
Use GitHub Copilot responsibly
Tempting but wrong
Typing only the keyword def on a blank line lets Copilot invent the best function unconstrained.
Why it fails
With no name, parameters, or stated intent, Copilot lacks the context it needs and is far more likely to produce an irrelevant or generic body. A descriptive signature and a comment describing inputs and output guide it toward the intended behaviour.
Improve developer productivity with GitHub Copilot
Tempting but wrong
Adding a file to .gitignore withholds it from Copilot context because untracked files are never sent as context.
Why it fails
.gitignore only controls what Git tracks; it has no effect on what Copilot reads as context. To stop a file being used by Copilot you must add its path to repository content exclusions, not to .gitignore.
Configure privacy, content exclusions, and safeguards
Tempting but wrong
Copilot uploads your entire local repository to GitHub on every keystroke so the model can index the whole project.
Why it fails
No. Copilot assembles a bounded prompt from the surrounding code and open-file context, not a full-repo upload per keystroke. Describing a complete repository transfer on each keystroke overstates what is actually transmitted.
Understand GitHub Copilot data and architecture
Tempting but wrong
Keeping a Copilot request to a single ambiguous word lets it infer the broadest, best interpretation of what you want.
Why it fails
A single ambiguous word strips out the goal, inputs, and constraints Copilot needs to scope the task, so it produces vague, unfocused output. State the goal, context, constraints, and expected output instead.
Apply prompt engineering and context crafting
Tempting but wrong
Edit Mode is appropriate when you want an explanation of code and explicitly no changes.
Why it fails
Wrong. Edit Mode is designed to propose and apply code changes for review. If the goal is to understand existing logic with nothing altered, it does too much. Copilot Chat answers in prose without modifying files.
Use GitHub Copilot features
Tempting but wrong
A Copilot suggestion referencing a helper that may not exist illustrates vendor lock-in.
Why it fails
Vendor lock-in is a commercial concern about being tied to one tool and finding it costly to switch later. It says nothing about whether a specific suggested helper actually exists or behaves correctly, which is the hallucination risk.
Use GitHub Copilot responsibly
Key terms
Exam-day rules
- Read the scenario for the need first. The task - explain without editing, draft a terminal command, withhold a file, steer output - is what picks the surface or control, so find it before you judge the options.
- Pick the surface by what must happen, not by which feature is newest. Pure explanation with no edits is Copilot Chat; a live shell command is the GitHub Copilot CLI; a scoped reviewable rewrite is Edit Mode; autonomous multi-step work is Agent Mode.
- Treat fluency as a trap, never a signal. Copilot produces confident phrasing whether or not the content is correct, so a convincing suggestion still has to be validated, and recent external facts must be confirmed against the authoritative source.
- Know what each control really does and reject the misuse. Content exclusions withhold named files as context, duplication detection suppresses public-code matches, the organisation Copilot policy is the central feature off switch - none of these makes output deterministic or guarantees a licence.
- Remember the developer stays accountable. Any option that treats an enabled filter, a clean run, or a successful compile as proof of correctness or a clear licence is the wrong answer; the human reviews and validates.
Revision schedule
- Day 1Map the blueprint and book a date
- Week 1Build the surface and control decision trees
- Weeks 1 to 2Go deep on features and responsible use
- Weeks 2 to 3Lock productivity and privacy controls
- Week 3Cover prompting and the data and architecture model