Session M2-02 lab: Agent mode in the IDE¶

Duration: 2 hours Difficulty: Intermediate Prerequisites: Sessions M1-01 to M1-03, M2-01 completed Deliverable: A tested Todo API built and extended with human review at each stage

Lab overview¶

You will build one Todo API throughout the lab. Scaffold it from a brief, then compare Ask and Agent mode on the same feature. Redirect the agent when the requirement changes. Finish with an accept, revise, or stop decision.

Stage	Work	Time
1	Prepare and scaffold the Todo API	30 min
2	Compare Ask and Agent mode on one feature	35 min
3	Redirect a multi-file change	35 min
4	Review the diff and test results, then finish the handoff	20 min

Before you start¶

Read the course safety baseline. Use a disposable workspace and synthetic data.

You need:

GitHub Copilot access in VS Code with Agent mode enabled;
Node.js and npm;
permission to let the agent read and edit the lab workspace;
permission to run the project test command;
lab/starter/project-brief.md, feature-spec.md, and course-correction.md.

If Copilot Agent mode is unavailable, stop the lab. Manual editing does not replace planning, tool approval, steering, and review in Agent mode.

Create a workspace outside the curriculum folder:

mkdir -p ~/copilot-labs/session-m2-02
cp sessions/session-m2-02-agent-mode-ide/lab/starter/project-brief.md \
  ~/copilot-labs/session-m2-02/
cp sessions/session-m2-02-agent-mode-ide/lab/starter/feature-spec.md \
  ~/copilot-labs/session-m2-02/
cp sessions/session-m2-02-agent-mode-ide/lab/starter/course-correction.md \
  ~/copilot-labs/session-m2-02/
cd ~/copilot-labs/session-m2-02

Create lab-notes.md with these headings:

# Session M2-02 Lab Notes

## Task contract
## Agent plan
## Tool and file review
## Ask versus Agent comparison
## Course correction
## Final decision

Stage 1: Prepare and scaffold the Todo API (30 minutes)¶

Write the task contract¶

Read project-brief.md. Add a short contract to lab-notes.md:

Goal: Build the Todo API described in project-brief.md.
Allowed scope: The session-m2-02 workspace only.
Commands: npm install, npm test, and npm start.
Non-goals: No database, front end, deployment, or extra dependencies.
Completion: Required routes exist and the test suite passes.
Stop: Pause on network access beyond npm, destructive commands, or work outside scope.

Ask Agent mode to plan first¶

Open the workspace in VS Code and select Agent mode. Use this prompt:

Read project-brief.md. Before editing, return a plan with the files you expect to
create, the commands you need, the tests that prove completion, and any decision
that is missing. Do not edit files yet.

Review the plan. Reject extra features and unexplained dependencies.

Scaffold the project¶

After accepting the plan, continue:

Create the Todo API in a todo-api directory. Stay within the approved plan. Pause
before installing packages. After approval, implement the project and run its tests.

At each pause:

Check the command and working directory.
Check that the action supports the brief.
Review changed files before the next command.
Record any revision in lab-notes.md.

Run the checks yourself:

cd todo-api
npm test

If the generated project has no working test script, require the agent to add one that covers create, list, validation, update, and not-found behavior.

Compare the shape of your project with lab/solution/todo-api/. The reference is an example, not a file-for-file requirement.

Checkpoint 1: baseline project works¶

The project lives in todo-api/.
The routes in project-brief.md exist.
The agent stayed inside the workspace.
You reviewed package changes before installation.
npm test passes.
lab-notes.md contains the accepted plan and any corrections.

Stage 2: Compare Ask and Agent mode on one feature (35 minutes)¶

feature-spec.md adds ownership and due-date behavior to the same Todo API.

Capture a clean baseline¶

Save the passing baseline:

cp -R todo-api todo-api-ask

Attempt the feature in Ask mode¶

Open todo-api-ask/, switch to Ask mode, and ask:

Read feature-spec.md and explain the exact code and tests needed to add this feature
to todo-api. Give changes file by file. Do not assume files that are not present.

Apply enough of the suggested changes to test the workflow. Track:

messages sent;
manual file edits;
copy and paste operations;
commands you ran;
missing context or incorrect assumptions.

Stop after 12 minutes. Save the result or diff. Keep the original todo-api/ as the clean baseline for the Agent mode attempt.

Implement the same feature in Agent mode¶

Switch to Agent mode:

Read feature-spec.md and inspect todo-api. Propose the smallest multi-file plan that
implements the feature without changing existing behavior. Include focused tests.
Wait for approval before editing.

Review the plan, approve only the work it describes, and watch each file and command. Then run:

cd todo-api
npm test

Add the comparison to lab-notes.md:

What you observed	Ask mode	Agent mode
Prompts or messages
Manual edits
Files changed
Commands run
Corrections needed
Final test result

Checkpoint 2: feature works and comparison is recorded¶

Existing tests still pass.
New tests cover owner and due-date behavior.
The implementation follows the existing project structure.
The Ask-versus-Agent table uses observed counts or examples.
You can explain why Agent mode did or did not help.

Stage 3: Redirect a multi-file change (35 minutes)¶

Open course-correction.md. Do not give it to the agent yet.

Start the original task¶

Ask Agent mode:

Refactor todo-api so route handlers delegate storage operations to a small repository
module. Keep the HTTP contract unchanged. Add or update tests, and stop after the
repository module and first route integration are complete.

Inspect the first stage:

Did the agent preserve current behavior?
Did it choose a clear module boundary?
Did it avoid unrelated formatting or dependency changes?
Do focused tests pass?

Apply the course correction¶

Now read course-correction.md and redirect the active task:

Pause. Apply the new constraints in course-correction.md. Revise the plan before
editing. Keep completed work only when it still fits the new requirement.

Reject any plan that needlessly rewrites the project. Check that the agent adjusts the repository boundary, validation, and tests while preserving the API.

Run the full suite:

cd todo-api
npm test

If the agent requests an unexpected dependency, network access, or broader file scope, choose stop. Record the unresolved work and complete the smallest safe change manually.

Checkpoint 3: redirect or stop decision is recorded¶

course-correction.md changed the plan before more edits.
The final diff stays inside todo-api/.
The API contract remains compatible.
Tests prove the changed behavior.
lab-notes.md records whether you accepted, revised, or stopped the run and why.

Stage 4: Review the diff and test results, then finish the handoff (20 minutes)¶

Review the final project as if another engineer submitted it.

Inspect git diff --stat and git diff, or use the editor's change view.
Check that each changed file meets a requirement in the brief, feature spec, or course correction.
Run npm test again.
Start the service and make one create request and one filtered list request.
Write the final decision: accept, revise, or stop.

Use lab/solution/final-review.md as the trainer reference. Review behavior, tests, and the diff rather than requiring an exact file layout.

Example requests:

curl -X POST http://localhost:3000/api/todos \
  -H "Content-Type: application/json" \
  -d '{"title":"Review agent work","owner":"learner","dueDate":"2026-10-15"}'

curl "http://localhost:3000/api/todos?owner=learner"

Do not approve the work only because the agent reports success. Use the diff, tests, and API behavior.

Checkpoint 4: another reviewer can reproduce the result¶

The final diff matches the accepted scope.
Package changes are explained.
Test output is saved in lab-notes.md.
The manual API check matches the feature spec.
The final decision names remaining risk.

Final deliverable¶

Submit the session-m2-02 workspace with:

The completed todo-api/ project.
Passing test output and two manual API checks.
lab-notes.md with the task contract, plan, tool and file review, comparison, course correction, and final decision.
A clear record of any stopped or rejected action.

Done: A reviewer can trace the final code to the approved plan, see where you redirected the agent, and reproduce the checks.