Literacy Check

20 questions, each with 4 choices. You may refer to the materials while you answer.

When you finish, click "Compile answers" at the bottom of the page. Your answers appear as text. Paste them into "Text to submit" on the submission form, write your name in "Name", and submit. The instructor will post the URL of the submission form in the Zoom chat.

The questions contain no client names or real figures. Do not write any in your answers either.

Which is the right way to use the 3 modes Ask, Agent and Plan in Copilot Chat?

They all have the same function. Only the place where they appear on the screen differs. Agent is the newest, so the right choice is to always use Agent for every task. Use Ask to have it read files and answer without changing them, Plan to have it lay out the steps of a task first, and Agent to have it edit files. Ask is only for questions, so it does not read the contents of the file open in the editor.

You want to make sure that Copilot Chat reads a specific file. Which method is appropriate?

In the chat, type # followed by the file name to attach the file, and check that it appears in the list of referenced files above the answer. If you write the file name in your message, its contents are read even without attaching it. The only way is to copy the whole file and paste it. A file that is not open in the editor cannot be read by any method.

Which statement about .github/copilot-instructions.md and AGENTS.md at the root of the workspace is correct?

AGENTS.md is not read unless you enable the chat.useAgentsMdFile setting. Both are personal settings files, so you put them in your home folder. Both are created automatically by an extension, so you must not edit their contents. Both are read by default. chat.useAgentsMdFile is a setting that you disable when you want to stop AGENTS.md from being read.

Which statement about the model assigned to a subagent is correct?

If you assign the subagent a higher-tier model than the parent, the parent's answers become more accurate. You cannot specify a model in a higher price tier than the parent's model. If you do, the subagent does not run, and you get back the models you can use. The subagent always uses the same model as the parent, and you cannot change it even in .agent.md. If you write a model name in the chat message and ask for "a subagent with that model", it always runs on that model.

Which statement correctly describes the difference between a prompt file (.prompt.md) and an agent skill (SKILL.md)?

A .prompt.md file saves a single prompt so that you can call it with /, and it is deprecated in Agent Host sessions. Its successor is SKILL.md, which bundles the steps and reference materials in a folder. SKILL.md is another name for .prompt.md, and both are written and stored in the same way. .prompt.md is the recommended method going forward, and SKILL.md is an experimental feature, so do not use it at work. SKILL.md is always loaded for every request, even if it has no description.

Which statement about test layers is correct?

The higher the layer, the more tests you should prepare. It is best to prepare the same number of tests for every layer. The idea of separating layers does not apply to teams that have not automated their tests. Lower layers have more tests that run quickly, and higher layers have fewer tests that take longer to run.

Which statement correctly describes how tests that "pass" relate to software that "meets" the specification?

If all the tests pass, you can say that the software meets the specification. Even if all the tests you wrote pass, the parts of the specification that you missed are not tested, so it is no guarantee that the software meets the specification. If coverage is 100%, you can say that the software meets the specification. If there are many tests, you can say that the software meets the specification.

When you have AI create test cases, what must a person always do?

Check only whether enough cases were generated. Delete only the cases that fail when you run them. Look for missing test viewpoints, point them out by name, and have the AI add them to the existing table. Adopt the generated result as it is, and judge by the run results.

Who decides the acceptance criteria? Which answer is appropriate?

Having AI read the specification and decide them automatically is the most accurate. All tests passing is itself the acceptance criteria. The person who implemented the feature can decide them after implementing it. People decide what counts as done. AI proposes draft criteria, and a person finalizes them.

Q10

What is the most appropriate purpose of unit tests?

To check that each piece of processing works as intended, including boundary values and invalid values To check that the screen layout is not broken To check a user's operations from start to finish To check that the processing speed meets the target

Q11

When you display text on the screen, which way of writing the code avoids XSS?

Put the received text into innerHTML to display it. Output the received text with document.write. Put the received text into textContent to display it. Embed the received text in the HTML as it is, and check it visually later.

Q12

Which is the appropriate way to think about input values?

If the screen checks the input, later processing can trust the values. Do not trust values that users enter. Before you use a value, convert it into a form suited to each place where it is used. For an internal company system, you can skip input checks. If you set a maximum length, you do not need to check the content.

Q13

When you review code generated by AI, which points should you look at most carefully?

Places that use values from outside as they are, and places where nothing happens when an error occurs Whether the variable names follow the company naming rules Whether there are enough comments Whether the code is kept short

Q14

Which is the correct cause of SQL injection?

The database does not have enough space. The communication is not encrypted. The query is built by joining the text that users entered as it is. The passwords are too short.

Q15

Which statement describes the place of E2E tests appropriately?

Checking all processing with E2E tests is the safest. They check the flow of user operations from start to end, so keep their number small and use them for important paths. They can be used instead of unit tests. The idea does not exist for systems that have no screen.

Q16

When you ask AI about your own work, which is the appropriate way to write the client?

Write the official name as it is. Use only the client's initial and write "Company A". Leave out the company name and write the system name instead. Generalize it to the level of industry and process.

Q17

Which statement about what you submit in this course is correct?

Submit the whole set of code you wrote. Submit only text in which you put your work into words. Do not submit code or real data. Take screenshots of the running screen and submit them. Attach the real data you used and submit it.

Q18

If a bug appears after AI-generated code goes into production, which is the correct way to think about responsibility?

The provider of the AI that generated the code is responsible. The developer did not write the code, so it is not the developer's responsibility. Only the reviewer is responsible. The side that adopted the code and released it is responsible.

Q19

When you start to use AI at work, which approach is appropriate?

Start with important deliverables to show results. Decide by yourself where it can be used, and do not share that with the people around you. Try it first on tasks with a small impact, check the output yourself, and then expand its use. Use it first for tasks that need especially high accuracy.

Q20

You notice that you entered information that you should have hidden. What should you do?

Report it promptly, following the company procedure. Deleting the conversation solves the problem, so do nothing. Send other questions to dilute the record. Decide to be careful from next time, and only keep a note of it.