Examine Copilot settings and interaction boundaries

Do not memorize the position of a button. Learn which control changes the role, which changes the runtime, and which changes permission to act.

Lab briefing

A small task module sits on a workbench with separate context trays and a permission control.

Original concept illustration (SVG)

Make a greeting trim whitespace without weakening input errors.

At a glance	Your route
Level and time	100; 35 minutes (facilitation estimate)
Starting action	Run the two original tests before asking Copilot to change the function.
Learner materials	Download 01-interface.zip
Workspace	Open the extracted kit root; run the baseline from `.` relative to that root
Expected initial check	The supplied baseline tests pass.
Setup help	Download, extract, local Git and optional GitHub

[!NOTE] Approving an answer is not the same as reviewing a change.

Concepts · First task · Evidence checklist · Reset

Learning objectives

Identify role, harness/target, model, context, and permissions independently.
Compare explanation, design, implementation, and inline suggestions.
Validate one small change without broadening authority.

Before you start

Complete Copilot setup and read the concepts reference. Prepare 01-interface using the work-drive procedure. The tiny Node fixture requires no package installation.

Concepts and use cases

Interaction	Appropriate use	Not proof of
Ask	Trace behavior and explain code	Correctness of a generated implementation
Plan	Define a safe change and checks	Files being changed
Agent	Implement with permitted tools	Every command requiring manual approval
Inline completion/NES	Accept or reject a local suggestion while editing	A complete repository refactor
Semantic rename	Rename one symbol across references	A change to its runtime behavior

Tool approval depends on your policy, not on whether a screenshot shows a Continue button. Only approve a command after reading its target, side effects, and scope.

Exercise scenario

The fixture formats a short welcome message. The requested change is to trim a name before formatting it while retaining the existing empty-name error.

Task 1 - Establish the starting behavior

Open the prepared workspace and inspect greeting.mjs and greeting.test.mjs.

Run:

node --test --test-concurrency=1 greeting.test.mjs

Record the output and exit status. A missing runtime is an environment failure, not a test of the greeting implementation.
Inspect the session target, role, model, and allowed tools. Keep unrelated MCP servers and background sessions out of this exercise.

Task 2 - Explain with and without explicit context

In Ask, request:

Explain how a blank name is handled. Do not edit or run commands.
Cite the file and branch that support your answer.

Record whether the agent retrieved files itself or lacked evidence. Do not assume that removing an attachment makes the file inaccessible.
Attach greeting.mjs explicitly and repeat. Compare correctness and cited evidence, not response length or confidence.
Confirm that the exercise request did not change files.

Task 3 - Plan a small, risky boundary change

Select Plan in a supported session.

Submit:

Trim the name before producing the greeting, without changing the error
for blank or non-string input. Identify whitespace and wrong-type cases,
affected files, the test command, and a one-change rollback plan. Do not edit.

Challenge any missing case: " Ada ", "", " ", null, and a number.
Approve only the bounded plan. Plan is useful even for a two-line change.

Task 4 - Implement and inspect

Hand the reviewed plan to Agent.
Allow only changes to the greeting implementation and its tests.
Review the diff before accepting it. Test weakening is not an acceptable fix.
Run the same test command yourself and add the missing whitespace case if it was not already covered.
Change one expected assertion deliberately, verify a nonzero result, then restore that assertion. This proves the test can reject wrong behavior.

Task 5 - Compare inline assistance

In a separate scratch file, begin writing a similar greeting function.
If inline suggestions are available, accept one and reject another. If none appears, record it; do not invent a suggestion.
Compare a manual symbol rename with next-edit suggestions.
Restore any language-specific preference you changed. Do not modify account- wide privacy settings or undocumented advanced engine URLs for this lab.

Verify your work

Baseline and after-change test outputs are recorded.
A negative assertion produces nonzero exit.
Role, target/harness, environment, model and permissions are not conflated.
Ask/Plan evidence is separate from the Agent diff.
No setting change leaks into other workspaces.

Troubleshooting

Use Command Palette names rather than screenshot coordinates. For missing context, inspect references and permissions. For a missing target, check policy and current client support. Old Ask/Edit screenshots are historical, not the current curriculum.

Independent practice

Repeat the change using a different available harness. Compare the actual tool list, instruction discovery, and validation output. Do not claim a speed improvement from a single trial.

Reset

Restore only the two exercise files from your disposable baseline, close the scratch file, and start a fresh session. Keep all original repository files untouched.