Performance Review

Purpose

code-review checks the performance of a single change as one dimension among several. That catches the N+1 query in the diff in front of you — but it never steps back to ask where the system actually spends its time, what the user actually waits on, or whether the architecture holds at 10× the data. Slowness that users feel is almost always systemic: an accumulation of individually-reasonable choices that no single diff review would flag.

This skill performs a measurement-driven performance review: it establishes what "slow" concretely means with a baseline measurement, then sweeps six dimensions of the system's performance posture — not line by line, but layer by layer. The output is a prioritized, severity-tiered findings report. Because it can be written in PRD format, the findings feed directly into design-to-tasks for remediation — the same handoff ui-design-audit and security-review use.

It is one of the suite's three audit skills, and shares their contract: standalone, whole-system, PRD-emitting. What ui-design-audit is to consistency and security-review is to posture, this skill is to speed.

When to use this vs. code-review