| name | code-review-performance |
| description | Run performance-focused code review when changes may affect latency, throughput, or CPU/memory/I/O efficiency on critical paths. Use for merge decisions requiring explicit performance-risk findings; do not use for broad non-performance review scope. |
Code Review Performance
Overview
Use this skill to detect performance regressions before merge, especially on hot paths and high-traffic execution flows.
Scope Boundaries
- Use this skill when the task matches the trigger condition described in
description.
- Do not use this skill when the primary task falls outside this skill's domain.
Inputs To Gather
- Hot-path endpoints/jobs and current performance budgets.
- Workload assumptions (QPS, payload size, concurrency, data cardinality).
- Existing benchmark/profiling evidence.
- Resource constraints (CPU, memory, I/O, network).
Deliverables
- Performance findings prioritized by user impact.
- Budget-fit judgment (within budget / at risk / out of budget).
- Required follow-up checks (benchmark, profiling, load test).
Finding Focus Areas
- Algorithmic growth (
O(n^2) regressions, repeated scans).
- Allocation pressure and unnecessary object churn.
- I/O amplification (N+1 calls, repeated DB/API access).
- Contention/serialization bottlenecks under concurrency.
- Cache invalidation or cache-bypass risks.
Quick Example
- Change adds per-item DB call inside loop over 10k records.
- Finding: high-severity throughput risk (N+1 query pattern).
- Fix direction: batch query + in-memory map, verify via benchmark.
Quality Standard
- Each finding ties to an explicit performance budget or hotspot.
- Recommendations include measurement plan, not assumptions only.
- Risk classification includes expected scale sensitivity.
- Missing evidence (benchmark/profile) is flagged explicitly.
Workflow
- Identify changed hot paths and performance-sensitive flows.
- Analyze algorithmic and resource behavior from diff.
- Compare expected behavior with existing budgets.
- Require measurement where uncertainty is material.
- Publish findings with mitigation and verification steps.
Failure Conditions
- Stop when high-risk regressions have no mitigation/verification plan.
- Escalate when performance impact cannot be bounded from available evidence.
