Pull request description template for AI-written changes

Save the block below as .github/pull_request_template.md (GitHub) or as a Markdown file in .gitlab/merge_request_templates/ (GitLab). Delete the sections your team doesn't use, and this note.

## What this changes

<!-- One or two sentences a reviewer can check the diff against. -->

## Why

<!-- The issue, ticket or decision this comes from. -->

Closes #

## How it was made

- [ ] Written by a person
- [ ] Written with an AI assistant (completion or chat)
- [ ] Written by a coding agent

Tool and model:
Task given to the tool (paste it, or link to it):

## What was verified

- [ ] Tests ran on this exact commit: <!-- link to the run -->
- [ ] New tests fail without this change
- [ ] No tests were deleted, skipped or loosened (or the reason is below)
- [ ] Checked by hand: <!-- what, and how -->

## What was NOT verified

<!-- Be specific. "Not run against production-sized data", "No browser test for Safari". Write "Nothing" only if that's true. -->

## Where to look first

<!-- The files or lines where a mistake would be expensive: money, auth, permissions, personal data, migrations. -->

## Out of scope

<!-- Anything the tool changed that the task didn't ask for, and why it stays. Empty is best. -->

## Risk and rollback

- Risk: low / medium / high, because
- Rollback: revert this commit / feature flag / other:
- Needs a migration or config change: yes / no

## Reviewer

- [ ] I read the task and the test results before the code
- [ ] I'm not the person who gave the tool its task

How to fill it in

What this changes. Write it so a reviewer can hold it next to the diff. If the summary and the diff disagree, one of them is wrong, and that is worth knowing before reading anything else.

How it was made. Tick one box and name the tool and model. Reviewers read agent-written code differently: the risk is less in the syntax and more in scope, assumptions and tests. Paste the task. It is the real specification of what the agent built.

What was verified. Each tick needs evidence behind it, ideally a link to the test run for this commit. A run from before the last push doesn't count. If you ticked "new tests fail without this change", you checked it by reverting the fix and running them.

What was NOT verified. This is the most useful section and the one most often left empty. Agents describe their own work as complete. A person filling this in is the check on that.

Where to look first. Point at the few places where an error costs real money or trust. A reviewer with limited time should spend it there.

Out of scope. Agents fix things they notice on the way. List them here, or better, move them to their own pull request.

Reviewer. Two boxes for the person approving. The second exists because the person who wrote the task is, in effect, the author of the change.