- Hands-on time
- About 30 minutes — provisional, from reading the steps, not from timing anyone.
- You will need
- A change produced by agent mode, the cloud agent or Copilot CLI, as a diff or a pull request
- The project's test command, and permission to run it
of 5 steps ticked
Read the change set, not the summary
Open the diff itself. An agent's summary describes what it meant to do; the diff is what it did. Note every file touched that you did not expect.
Review the change set, not the summary — the lesson section this comes from
Before you tick it: A list of touched files, with any unexpected ones marked.
Check the diff against the stated requirement
Take the original request and check each part of it against the diff. Look for what is missing as hard as for what is wrong — agents finish confidently.
Reviewing an agent diff — the lesson section this comes from
Before you tick it: Each requirement marked met, partly met, or absent.
Run the tests, or write down that you did not
Run the project's test command against the change. If you cannot — no environment, no time — record which tests were not run. An unrun test is not a passed test.
Validation — the lesson section this comes from
Before you tick it: Test output with pass/fail counts, or an explicit list of tests not run.
Look for the changes agents get wrong
Check for new dependencies, changed permissions, secrets in code, weakened validation, and tests edited to pass rather than code fixed to pass.
Reviewing agent-authored changes — the lesson section this comes from
Before you tick it: Each of the five categories checked, with findings or 'none'.
Decide, and write it down
Merge, request changes, or discard — and one line on what still needs attention if you merge. A merge with an open question is a merge with a follow-up issue.
Reading review output well — the lesson section this comes from
Before you tick it: A decision and, if merging, the follow-up written where it will be seen.
The finish lineFinished
You can show the diff you read, the test results or the list of unrun tests, and the decision with its open items.
Ticks are your own record that you did each step. They are kept in this browser under one key and do not mark the source lessons complete — a step is one section of a lesson, not the lesson.
One next step
You reviewed someone else's agent change. The lab has you scope, run and review one of your own, with a validation step that can fail.
Only this session. Lessons, labs and other sessions are untouched.