New Feature / Enhancement Checklist
Current Limitation
The Tasks checklist in the PR template covers tests, docs, security checks and error codes, but nothing asks authors to confirm they've reviewed and understood their own change. With AI agents producing more of the code in PRs, it's become easy to submit a working-looking change the author can't actually explain, which shifts the whole comprehension burden onto reviewers.
Feature / Enhancement Description
Add a self-review item to the Tasks checklist in .github/pull_request_template.md, something like:
- [ ] I have reviewed this PR line by line and understand how the change works and why it is correct
Exact wording open for discussion - the point is that ticking it is the author signing that the PR isn't unreviewed AI output.
Example Use Case
- An author (or their coding agent) prepares a change
- Before opening the PR, the author reads the diff and makes sure they can explain it
- They tick the checkbox
- Reviewers can ask the author to walk through any part of the change, and the checkbox sets that expectation up front
Alternatives / Workarounds
Reviewers asking authors to explain their changes case by case. Works, but sets no expectation up front and usually happens after review time has already been spent.
3rd Party References
Projects like curl and QEMU have introduced contribution policies requiring authors to disclose AI assistance and vouch that they understand and verified what they're submitting.
New Feature / Enhancement Checklist
Current Limitation
The Tasks checklist in the PR template covers tests, docs, security checks and error codes, but nothing asks authors to confirm they've reviewed and understood their own change. With AI agents producing more of the code in PRs, it's become easy to submit a working-looking change the author can't actually explain, which shifts the whole comprehension burden onto reviewers.
Feature / Enhancement Description
Add a self-review item to the Tasks checklist in
.github/pull_request_template.md, something like:Exact wording open for discussion - the point is that ticking it is the author signing that the PR isn't unreviewed AI output.
Example Use Case
Alternatives / Workarounds
Reviewers asking authors to explain their changes case by case. Works, but sets no expectation up front and usually happens after review time has already been spent.
3rd Party References
Projects like curl and QEMU have introduced contribution policies requiring authors to disclose AI assistance and vouch that they understand and verified what they're submitting.