Skip to content

How to review a task ​

Review is where the team's quality is truly at stake: it's the last look before the merge. Two distinct processes: reviewing a feature ticket in the product team, and peer review between QAs.

Reviewing a feature ticket ​

When development finishes (refined, developed, code-reviewed), the ticket moves to "To Review" and any team member — dev or QA — can review it. My flow:

  1. Assign the ticket to yourself before starting — make it visible who's reviewing what.
  2. Understand the purpose of the task. Reviewing without context is looking, not reviewing.
  3. Check the branch's pipeline: development tests and QA tests. If the development ones fail, back to the developer — QA shouldn't spend time detecting what the automated gates already detect. Corollary: don't assign QA a ticket with a red pipeline.
  4. Run the test cases proposed by the developer and add your own if you spot coverage gaps. As a QA, here I also review two more things: that the acceptance criteria are covered by the developer's tests, and whether any functional test of mine has become redundant because a development test already covers it — removing duplication is reviewing too.
  5. Problems? Comment on the ticket specifying the error (not "it doesn't work") and re-test the fix.
  6. Closing: once the ticket is validated, if there are no conflicts the reviewer merges; if there are, the reviewer records that it's validated and the developer resolves the conflicts and merges.
  7. On delivery, fill in the fix version: knowing which version each change ships in is not optional.

Peer review between QAs ​

The QA team's own tasks (automation framework, pipelines, documentation) also go through peer review — every task, always, even when there are default reviewers.

The author prepares the review ​

Before putting the ticket in the review column (unassigned, so pending ones stay visible):

  • Pipeline run on their branch — if the task includes tests, the evidence that they work goes up front.
  • Link to the case/suite in the test case manager that it covers — the reviewer understands what is being covered.
  • Any needed documentation or context, in the ticket.
  • A heads-up in the team channel.

The reviewer ​

  1. Self-assigns and reacts to the channel message (visibility).
  2. Checks three things: the code follows good practices; the pipeline passes; and — the one that gets forgotten — the changes solve the problem described in the ticket. Recommendation: pull the changes down and try them locally.
  3. Changes needed? Native comments on the PR + "Changes requested" + reassign to the author. On complex tasks, a pair review speeds things up.
  4. Closing: the reviewer approves and merges, fills in version and components, moves it to "Delivered" and unassigns themselves.

The golden rule ​

Pending reviews take priority over starting new work. Unblocking a teammate multiplies the team's throughput; your own task only speeds you up.

Key idea

A review isn't a formality before the merge: it's the check that the change solves the problem in the ticket, with evidence. A green pipeline tells you the tests pass — only the reviewer can tell you they're the right tests.