subagent-creation_skill

Overview

This skill creates and configures Claude Code subagents for focused task delegation. It helps define reusable expert assistants with custom prompts, tool permissions, and model choices to automate code review, debugging, testing, performance analysis, and documentation tasks. Use it to streamline team workflows and enforce consistent agent behavior.

How this skill works

Subagents are defined as Markdown files with YAML frontmatter placed in project or user agent directories. Each agent file specifies name, description, tools, and model; the body contains the system prompt, checklists, and output format. Claude delegates tasks to these isolated, single-responsibility agents based on explicit invocation or automatic delegation rules.

When to use it

• Set up automated code review and security checks after commits
• Create a dedicated debugger agent to diagnose and fix failing tests
• Generate unit and integration tests via a test-writer subagent
• Run performance analysis and produce optimization reports
• Standardize documentation updates using a doc-writer agent

Best practices

• Give each agent a single responsibility and clear role statement
• Grant minimal tool permissions needed (Read, Write, Edit, Grep, Glob, Bash, LSP)
• Include a concrete checklist and exact output template in the prompt
• Prefer model: inherit unless task needs sonnet/haiku/opus behavior
• Store agents in .claude/agents for project sharing and version control
• Test prompts iteratively and refine based on real runs

Example use cases

• Post-commit hook invokes code-reviewer to scan the latest commit for security issues
• Chain performance-analyzer then debugger to find and fix a bottleneck
• Invoke test-writer to produce unit tests for newly added payment processing code
• Resume a paused code-reviewer session to continue a multi-step review
• Create a database-expert agent to optimize queries and suggest schema changes