开发与测试
powershell-performance-review
Find and fix performance problems in PowerShell code by measuring it. Use when asked to optimize, speed up, profile, or analyze the performance of PowerShell code.
fadwen · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add fadwen/ai-powershell-standards --skill powershell-performance-review技能说明
name: powershell-performance-review description: Find and fix performance problems in PowerShell code by measuring it. Use when asked to optimize, speed up, profile, or analyze the performance of PowerShell code.
Performance review of PowerShell code
The collection and string rules are in .github/instructions/powershell-style.instructions.md
(STY-22, STY-40 to STY-45, and STY-50). Read them first, unless that file is already loaded.
Steps
- Ask what is slow and how large the real input is, if the user has not said. An improvement that matters at 100,000 items can be noise at 100.
- Measure before changing anything. Use
Measure-Commandaround the code with realistic input, and repeat the run several times. If you cannot run the code, say so and present your suggestions as unmeasured. - Look for the causes that usually dominate:
- An array grown with
+=inside a loop, or a string extended with+=about 100 times or more. Below that,+=on a string is fine (STY-22); do not report it. - A command called once for each item when it accepts all items at once, or a remote or disk call inside a loop that could be made once outside it.
- Filtering with
Where-Objectwhat the source command could have filtered. - A lookup that scans a collection inside a loop, where a hashtable built once would do.
ForEach-Objectover a large collection that is already in memory.- Output piped to
Out-Nullin a tight loop.
- An array grown with
- Change one thing at a time and measure again with the same input. Keep a change only if it is faster and the output is the same.
- Do not trade correctness or clarity for a small gain, and do not add parallelism
(
ForEach-Object -Parallel, thread jobs) unless the work is slow per item and the items are independent.
Report
Give the measurements before and after, with the input size and the PowerShell version, for each change you kept. Performance figures MUST come from runs you made; do not estimate percentages. List what you tried that did not help.