review-performance
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseSkill: Review Performance
Skill：性能评审
Purpose
目的
Review code for performance concerns only. Do not define scope (diff vs codebase) or perform security/architecture/language-framework convention analysis; those are handled by other atomic skills. Emit a findings list in the standard format for aggregation. Focus on algorithmic complexity, query efficiency, I/O and network cost, memory behavior, contention and concurrency bottlenecks, caching strategy, and measurable regression risk.
仅针对代码的性能问题进行评审。无需定义评审范围（如diff对比或整个代码库），也不进行安全/架构/语言框架规范分析；这些内容由其他原子技能负责。按照标准格式输出问题发现列表，以便进行汇总。重点关注算法复杂度、查询效率、I/O与网络成本、内存行为、竞争与并发瓶颈、缓存策略，以及可量化的退化风险。
Use Cases
适用场景
- Orchestrated review: Used as a cognitive step when review-code runs scope -> language -> framework -> library -> cognitive.
- Performance-focused review: When the user wants only performance dimensions checked before merge or release.
- Regression prevention: Validate that changes do not introduce obvious latency, throughput, or memory regressions.
When to use: When the task includes performance review. Scope and code range are determined by the caller or user.
- 编排式评审：在review-code执行范围→语言→框架→库→认知层面的评审流程时，作为认知步骤使用。
- 聚焦性能的评审：当用户希望在合并或发布前仅检查性能维度时使用。
- 退化预防：验证代码变更是否引入明显的延迟、吞吐量或内存退化问题。
使用时机：当任务包含性能评审需求时。评审范围和代码范围由调用方或用户确定。
Behavior
行为规范
Scope of this skill
本技能的评审范围
- Analyze: Performance dimensions in the given code scope (files or diff provided by the caller). Do not decide scope; accept the code range as input.
- Do not: Perform scope selection, security review, architecture review, or language/framework style review. Focus only on performance.
- 分析内容：在给定的代码范围（调用方提供的文件或diff）内的性能维度问题。无需自行确定范围，直接接受提供的代码范围作为输入。
- 禁止操作：不得进行范围选择、安全评审、架构评审或语言/框架风格评审。仅聚焦性能问题。
Review checklist (performance dimension only)
评审检查清单（仅性能维度）
- Complexity hotspots: Detect unnecessary O(n^2)+ behavior, repeated scans, nested loops over large collections, and avoidable recomputation.
- Database and query efficiency: N+1 access patterns, missing pagination, broad selects, inefficient joins/filters, and query frequency amplification.
- I/O and network cost: Chatty remote calls, missing batching, blocking calls on critical paths, unbounded retries/timeouts, and poor backoff behavior.
- Memory and allocations: Excessive allocations/churn, large object retention, unnecessary copies, unbounded growth, and avoidable buffering.
- Concurrency and contention: Lock contention, serialized critical sections, thread/goroutine starvation, queue backpressure, and oversubscription risks.
- Caching and reuse: Missing cache opportunities on hot read paths, invalidation correctness risks, stampede risk, and low-value cache layers.
- Load-facing behavior: Missing limits/guards (batch size, page size, concurrency caps), expensive defaults, and absent degradation strategy under load.
- Observability for performance: Missing metrics/tracing around hot paths that prevents regression detection and capacity planning.
- 复杂度热点：检测不必要的O(n^2)+级行为、重复扫描、针对大型集合的嵌套循环，以及可避免的重复计算。
- 数据库与查询效率：N+1访问模式、缺失分页、宽泛的查询语句、低效的关联/过滤逻辑，以及查询频率放大问题。
- I/O与网络成本：频繁的远程调用、缺失批量处理、关键路径上的阻塞调用、无限制的重试/超时，以及不合理的退避策略。
- 内存与分配：过度的内存分配/周转、大对象滞留、不必要的复制、无限制的内存增长，以及可避免的缓冲操作。
- 并发与竞争：锁竞争、序列化的关键代码段、线程/协程饥饿、队列背压，以及过度订阅风险。
- 缓存与复用：热点读路径上缺失缓存机会、缓存失效的正确性风险、缓存击穿风险，以及低价值的缓存层。
- 面向负载的行为：缺失限制/防护机制（如批量大小、分页大小、并发上限）、开销高昂的默认配置，以及负载下的降级策略缺失。
- 性能可观测性：热点路径上缺失指标/追踪，导致无法检测退化问题和进行容量规划。
Severity guidance
严重程度指南
- critical: Clear production impact likely (e.g. unbounded loop/growth, repeated expensive I/O in hot path, catastrophic query pattern).
- major: Strong regression or scalability risk with realistic traffic/data growth.
- minor/suggestion: Localized or lower-impact optimization opportunities.
- critical（严重）：可能对生产环境造成明确影响（如无限循环/内存增长、热点路径中重复执行的高开销I/O、灾难性查询模式）。
- major（主要）：随着流量/数据增长，存在明显的退化或可扩展性风险。
- minor/suggestion（次要/建议）：局部或低影响的优化机会。
Tone and references
语气与参考要求
- Professional and technical: Reference specific locations (file:line or query/block).
- Emit findings with Location, Category, Severity, Title, Description, Suggestion.
- 专业且技术化：引用具体位置（如文件:行号 或 查询/代码块标识）。
- 输出的问题发现需包含位置、类别、严重程度、标题、描述、建议。
Input & Output
输入与输出
Input
输入
- Code scope: Files or directories (or diff) already selected by the user or scope skill. This skill does not decide scope; it reviews the provided code for performance only.
- 代码范围：由用户或范围技能已选定的文件、目录（或diff）。本技能不负责确定范围，仅针对提供的代码进行性能评审。
Output
输出
- Emit zero or more findings in the format defined in Appendix: Output contract.
- Category for this skill is cognitive-performance.
- 输出零个或多个问题发现，格式需符合附录：输出约定中的定义。
- 本技能对应的类别为cognitive-performance。
Restrictions
限制条件
- Do not perform scope selection, security, architecture, or language/framework style review. Stay within performance dimensions.
- Do not give conclusions without specific locations or actionable suggestions.
- Do not claim benchmark numbers unless measured evidence is provided in the input.
- 禁止进行范围选择、安全、架构或语言/框架风格评审。仅聚焦性能维度。
- 禁止在未提供具体位置或可执行建议的情况下给出结论。
- 禁止在输入中未提供实测证据的情况下声称基准测试数据。
Self-Check
自我检查
- Was only the performance dimension reviewed (no scope/security/architecture/style)?
- Are complexity, query efficiency, I/O, memory, concurrency, caching, and load behavior covered where relevant?
- Is each finding emitted with Location, Category=cognitive-performance, Severity, Title, Description, and optional Suggestion?
- Are high-impact regression risks clearly distinguished from minor optimizations?
- 是否仅评审了性能维度（未涉及范围/安全/架构/风格）？
- 是否涵盖了复杂度、查询效率、I/O、内存、并发、缓存及负载行为（相关情况下）？
- 每个问题发现是否都包含位置、类别=cognitive-performance、严重程度、标题、描述，以及可选的建议？
- 是否明确区分了高影响的退化风险与次要优化建议？
Examples
示例
Example 1: N+1 query pattern
示例1：N+1查询模式
- Input: Loop fetches child records per parent with one query per iteration.
- Expected: Emit a major/critical finding for N+1 behavior; suggest batch query or join strategy. Category = cognitive-performance.
- 输入：循环中针对每个父记录单独查询子记录，每次迭代执行一次查询。
- 预期输出：针对N+1行为输出major或critical级别的问题发现；建议采用批量查询或关联策略。类别为cognitive-performance。
Example 2: Hot-path allocation churn
示例2：热点路径的内存分配周转
- Input: Request handler repeatedly allocates large temporary buffers and serializes payload multiple times.
- Expected: Emit a major finding for allocation pressure and latency impact; suggest reuse/pooling or single-pass transform. Category = cognitive-performance.
- 输入：请求处理程序重复分配大型临时缓冲区，并多次序列化负载。
- 预期输出：针对内存分配压力和延迟影响输出major级别的问题发现；建议采用复用/池化或单遍转换策略。类别为cognitive-performance。
Edge case: No clear performance risk in small formatting diff
边缘案例：小型格式调整diff中无明确性能风险
- Input: Diff includes comments/renaming only, no behavioral changes.
- Expected: Emit no findings or one suggestion-level note; do not invent optimization work. Category remains cognitive-performance for any emitted finding.
- 输入：diff仅包含注释/重命名，无行为变更。
- 预期输出：不输出任何问题发现，或输出一条建议级别的说明；不得凭空提出优化需求。若输出问题发现，类别仍为cognitive-performance。
Appendix: Output contract
附录：输出约定
Each finding MUST follow the standard findings format:
| Element | Requirement |
|---|---|
| Location | |
| Category | |
| Severity | |
| Title | Short one-line summary. |
| Description | 1-3 sentences. |
| Suggestion | Concrete fix or improvement (optional). |
Example:
markdown
- **Location**: `service/orders/handler.go:118`
- **Category**: cognitive-performance
- **Severity**: major
- **Title**: Per-item remote call inside request loop
- **Description**: The handler performs one downstream call per item, creating linear remote round-trips and latency growth.
- **Suggestion**: Batch requests or prefetch related data once per request; add timeout and bulk size guards.每个问题发现必须遵循标准格式：
| 元素 | 要求 |
|---|---|
| Location（位置） | |
| Category（类别） | |
| Severity（严重程度） | |
| Title（标题） | 简短的单行摘要。 |
| Description（描述） | 1-3句话。 |
| Suggestion（建议） | 具体的修复或改进方案（可选）。 |
示例：
markdown
- **Location**: `service/orders/handler.go:118`
- **Category**: cognitive-performance
- **Severity**: major
- **Title**: Per-item remote call inside request loop
- **Description**: The handler performs one downstream call per item, creating linear remote round-trips and latency growth.
- **Suggestion**: Batch requests or prefetch related data once per request; add timeout and bulk size guards.