pr-review

/octopus:pr-review is the agent’s self-review pass on a PR it authored (or assisted). It runs after pr-open and before human review: pulls the diff, walks a six-axis checklist, reports findings to the author, and — once any blocking issues are addressed — assigns the human reviewers configured in .octopus.yml.

Why self-review before assigning reviewers

The cheapest fixes are the ones reviewers never see. The six-axis checklist catches the obvious classes of mistake before they consume reviewer attention:

• Correctness — does the code do what the PR description claims?
• Design — does the change fit the existing architecture, or does it introduce a new pattern?
• Readability — can a reviewer who didn’t write this code understand it on first read?
• Edge cases — null checks, empty inputs, race conditions, off-by-one boundaries.
• Security — injection, auth bypass, secrets, data exposure.
• Tests — are the right things being tested, at the right level?

The self-review is not adversarial — it’s the same pass a disciplined contributor would do before hitting “Ready for review”. Codifying it as a command means it happens consistently instead of being skipped under time pressure.

What pr-review is not

It’s a smaller scope than /octopus:codereview. codereview is a comprehensive security and quality review of uncommitted changes — wider in scope (audits secrets, scans for leftover debug statements, verifies tests run) and meant to be invoked deliberately before pushing.

pr-review is faster and PR-focused — it’s the discipline that runs every time, on every PR, with no skip.

Reviewer assignment

Reviewers come from .octopus.yml:

reviewers:
  - alice
  - bob

If the list is empty, the command leaves the PR unassigned and informs the user. Reviewer rotation (round-robin, area expertise, etc.) is out of scope — the manifest is the configuration surface.

Pairing

/octopus:pr-comments is the next stop once feedback arrives — it walks the receiving code-review discipline (verify the critique, ask for evidence on generic comments, separate reasoned feedback from preference, never make performative changes).

Source: commands/pr-review.md