Use this skill when asked to review a task you did or independently after a bigger change. The reviewer must run on GPT-5.4 with xhigh reasoning, receive a concise neutral summary of the task and implemented solution, and check for bugs, regressions, and missing tests without being primed toward a positive conclusion.
Run this skill after larger tasks or changes. A meaningful change is any non-trivial behavior change, refactor, or update that materially affects the implementation. Do not run the reviewer for small fixes or immediately after addressing the last review comment.
Use a separate review agent to challenge the latest change before handoff. The review should be independent and skeptical, with the primary goal of finding bugs, regressions, weak assumptions, and missing coverage. Reviewing takes time. Make sure to let the agent finish its work, don't abort it early.
model: gpt-5.4 and reasoning_effort: xhigh.The reviewer should receive a compact, neutral summary. Good inputs state what changed and what constraints matter. Bad inputs sell the solution.
Use a prompt in this shape:
Review this change independently. Look for bugs, behavioral regressions, invalid assumptions, edge cases, and missing tests. Also ask whether the refactor fits in the overall architectural direction, including for future work and scaling.
Task summary:
- ...
Constraints:
- ...
Implemented change summary:
- ...
Changed files:
- ...
Avoid prompts in this shape:
I fixed the issue cleanly. Please sanity-check the solution.
The point of the review agent is to pressure-test the work, not to confirm a positively framed story.