Ensure accessibility in UI components including semantic HTML, ARIA attributes, keyboard navigation, and WCAG 2.2 AA compliance.
日本語の概要は準備中です。原文の説明を表示しています。
PowerShell 7+ scripting and Windows system administration -- cross-platform automation, secure credential handling, Pester testing, DSC, and JEA patterns.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
This skill covers PowerShell 7+ (Core) for cross-platform automation, system administration, and DevOps scripting. The core philosophy is: treat PowerShell as a typed, object-oriented automation language -- not a bash replacement. Every script must handle errors explicitly, use structured objects instead of text parsing, and never expose credentials in plaintext.
$ErrorActionPreference = 'Stop' at the top of scripts -- silent failures are the primary cause of automation bugs and data loss.Microsoft.PowerShell.SecretManagement module to pull secrets from vaults.[PSCustomObject] or -OutputType JSON for structured output -- text parsing with regex is fragile and breaks on locale/format changes.Invoke-Expression (IEX) on untrusted input -- it is the PowerShell equivalent of eval() and enables arbitrary code execution.| Anti-Pattern | Why It Fails | Correct Approach |
|---|---|---|
| Parsing command output with regex instead of using objects | Breaks on locale changes, format updates, and different OS versions | Use cmdlet object output directly or convert to PSCustomObject |
Using Invoke-Expression to build dynamic commands | Enables code injection; any user input can execute arbitrary PowerShell | Use splatting (@params) for dynamic parameters; use Start-Process for external commands |
| Catching all exceptions with empty catch blocks | Silently swallows errors; automation appears to succeed when it failed | Use typed catch blocks; log and re-throw unexpected exceptions |
| Using Windows PowerShell 5.1 syntax without checking compatibility | Scripts fail on Linux/macOS where PS 7 is the only option | Use $PSVersionTable.PSVersion checks; prefer PS 7 cross-platform cmdlets |
| Storing credentials in script variables or config files | Plaintext secrets in source control; credential theft risk | Use Get-Secret from SecretManagement module; inject via environment variables in CI |
#Requires -Version 7.0
#Requires -Modules @{ ModuleName='Microsoft.PowerShell.SecretManagement'; ModuleVersion='1.1.0' }
$ErrorActionPreference = 'Stop'
Set-StrictMode -Version Latest
function Invoke-DataBackup {
[CmdletBinding()]
param(
[Parameter(Mandatory)]
[ValidateNotNullOrEmpty()]
[string]$TargetPath,
[Parameter()]
[switch]$Compress
)
begin {
Write-Verbose "Starting backup to $TargetPath"
}
process {
try {
# Business logic here
}
catch [System.IO.IOException] {
Write-Error "IO error during backup: $_"
throw
}
catch {
Write-Error "Unexpected error: $_"
throw
}
}
end {
Write-Verbose "Backup complete"
}
}
# Register a vault (one-time setup)
Register-SecretVault -Name 'AzureKeyVault' -ModuleName 'Az.KeyVault'
# Retrieve secret at runtime
$apiKey = Get-Secret -Name 'MyApiKey' -Vault 'AzureKeyVault' -AsPlainText
# Use in automation (never log the value)
$headers = @{ 'Authorization' = "Bearer $apiKey" }
Invoke-RestMethod -Uri $endpoint -Headers $headers
# Process structured data through the pipeline
Get-ChildItem -Path $target -Filter *.json |
ForEach-Object {
$data = Get-Content -Path $_.FullName | ConvertFrom-Json
[PSCustomObject]@{
FileName = $_.Name
ItemCount = $data.items.Count
LastModified = $_.LastWriteTime
}
} |
Sort-Object -Property ItemCount -Descending |
Export-Csv -Path 'report.csv' -NoTypeInformation
# Invoke-DataBackup.Tests.ps1
Describe 'Invoke-DataBackup' {
BeforeAll {
. $PSScriptRoot/Invoke-DataBackup.ps1
}
Context 'When target path exists' {
It 'Should create backup file' {
$result = Invoke-DataBackup -TargetPath $TestDrive
$result | Should -Not -BeNullOrEmpty
Test-Path "$TestDrive/backup.zip" | Should -BeTrue
}
}
Context 'When target path is invalid' {
It 'Should throw IO exception' {
{ Invoke-DataBackup -TargetPath '/nonexistent/path' } |
Should -Throw -ExceptionType ([System.IO.IOException])
}
}
}
# Use Join-Path for all path operations
$configPath = Join-Path -Path $HOME -ChildPath '.config' -AdditionalChildPath 'myapp', 'settings.json'
# Check platform before using platform-specific features
if ($IsWindows) {
# Windows-specific: registry, WMI, COM
$os = Get-CimInstance -ClassName Win32_OperatingSystem
} elseif ($IsLinux -or $IsMacOS) {
# Unix-specific: /proc, systemctl
$os = uname -a
}
| Skill | Relationship |
|---|---|
devops | CI/CD pipeline integration with PowerShell scripts |
docker-compose | Containerized PowerShell automation |
terraform-infra | Infrastructure provisioning alongside PS configuration |
tdd | Test-driven development methodology for Pester tests |
Before starting:
Read .claude/context/memory/learnings.md for prior PowerShell modules, Pester testing patterns, or OS-specific workarounds.
After completing: Record new PowerShell modules, Pester testing patterns, or OS-specific workarounds to .claude/context/memory/learnings.md.
ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Ensure accessibility in UI components including semantic HTML, ARIA attributes, keyboard navigation, and WCAG 2.2 AA compliance.
日本語の概要は準備中です。原文の説明を表示しています。
Use when you want to improve response quality through meta-cognitive reasoning. Applies 15+ reasoning methods to reconsider and refine initial outputs.
日本語の概要は準備中です。原文の説明を表示しています。
N-round opposing-stance debates for trade-off analysis. Assigns pro/con roles to agents, runs structured debate rounds with quality scoring, and produces a moderator synthesis with confidence-rated recommendation. Generalizable to architecture, technology, security, and design decisions.
日本語の概要は準備中です。原文の説明を表示しています。
Force adversarial code review stance that eliminates confirmation bias — reviewer must find issues or re-analyze
日本語の概要は準備中です。原文の説明を表示しています。
Creates specialized AI agents on-demand when no existing agent matches a request. Use when the Router cannot find a suitable agent for a task. Enables self-evolution by generating persistent agents.
日本語の概要は準備中です。原文の説明を表示しています。
LLM-as-judge evaluation framework with 5-dimension rubric (accuracy, groundedness, coherence, completeness, helpfulness) for scoring AI-generated content quality with weighted composite scores and evidence citations
日本語の概要は準備中です。原文の説明を表示しています。