AgentSkillsCN

command-management

在需要按照最佳实践创建或更新自定义命令时，主动使用此技能

SKILL.md
--- frontmatter
name: command-management
description: Use PROACTIVELY this skill when you need to create or update custom commands following best practices
allowed-tools: Read, Write, SlashCommand, TodoWrite, Glob, Grep

Goal: Create or update a custom command following best practices and template standards

Tasks

  • •T001: Read the Anthropic documentation on how to create custom commands from @.claude/skills/command-management/references/command-docs.md
  • •T001: Read the command template from .claude/skills/command-management/templates/command.md
  • •T002: Analyze instructions/requirements from the user to understand the command's purpose and scope
  • •T003: Determine the appropriate command folder location
  • •T004: Create or update the command file in the determined location
  • •T005: Test the command using the SlashCommand tool
  • •T006: Report the results to the user

Constraints

  • •NEVER omit any of the 6 template sections (Context, Tasks, Constraints, Examples, References, Output Format)
  • •NEVER omit required YAML frontmatter fields (name, description, allowed-tools, argument-hint, model)
  • •NEVER save command files outside .claude/commands/ directory
  • •NEVER grant more tool permissions than absolutely necessary
  • •NEVER use 'all' tool access unless the command truly requires unrestricted access
  • •NEVER use 'opus' model unless the command requires complex reasoning
  • •DO NOT create tasks without sequential IDs (T001, T002, T003, etc.)
  • •DO NOT create non-atomic tasks - each task must describe one clear operation
  • •DO NOT skip the testing phase via SlashCommand tool
  • •DO NOT omit usage examples with actual parameter values

Success Criteria

  • • Command file saved to correct location (.claude/commands/ or appropriate subfolder)
  • • YAML frontmatter includes all required fields (name, description, allowed-tools, argument-hint, model)
  • • All 6 template sections present and populated
  • • Tool permissions follow least-privilege principle
  • • Tasks organized with sequential IDs (T001, T002, etc.)
  • • At least one concrete usage example included
  • • Command tested successfully via SlashCommand tool