AI review

Agent review is automatic. When an issue reaches AI Review (the working agent created a PR), Cate dispatches it to an idle reviewer tab. You don’t need to do anything.

The two-pass structure

The reviewing agent uses a mandatory two-pass structure:

Pass 1 — Spec compliance: The reviewer walks every acceptance criterion from the issue against the PR diff. Each criterion gets an explicit verdict: Met, Partially Met, or Not Met. If any criterion is not met, the review stops immediately — there is no Pass 2.

Pass 2 — Code quality: Only after all acceptance criteria pass does the reviewer evaluate code quality. Findings are categorized as Critical (blocking), Important (should fix), or Suggestion (optional). Only Critical findings block the review.

If the issue included a UI mockup in its body, the agent compares the implementation against the agreed layout and flags structural drift. Layout mismatches are treated as acceptance criteria failures.

Outcomes

No blocking issues: The agent posts a review comment summarizing what it verified and moves the issue to In Review. The PR is now waiting for you.

Blocking issues found: The agent posts line-specific change requests with file and line references for each issue. The issue moves back to To Do. The original working agent picks it up — change request items get priority — addresses the feedback, and resubmits.

Epic subtask PRs

For sub-issues that are part of an epic, the PR targets the epic branch instead of main. When the reviewing agent finds no blocking issues:

• cate label: The reviewer squash-merges the PR to the epic branch and closes the sub-issue automatically. No human review step for individual sub-issues.
• cate:supervised label: The reviewer posts a comment asking you to merge the PR. The sub-issue moves to In Review and waits for you.

When all sub-issues are merged to the epic branch, the tracking PR is automatically marked ready and the epic moves to In Review.