subagent-builder

Design a Claude Subagent by defining its task scope, writing a focused system prompt that specifies role and output format, selecting appropriate tools (Read, Write, Edit, Bash, Grep, Glob), and choosing the right model (Haiku, Sonnet, Opus). This Skill guides you through each step to ensure isolation and controlled execution.

For code analysis and security scanning, equip your Subagent with Read for file access, Grep for pattern matching, Bash for execution, and Edit for modifications. Select tools based on your Subagent's specific task scope to maintain isolation while enabling necessary operations.

Validate your Subagent by testing its system prompt against sample inputs, verifying tool access works as expected, confirming context isolation functions correctly, and running integration tests with your main workflow. This Skill provides validation protocols to catch issues before production use.

Create a Subagent when you need isolated context, specialized instructions, or controlled tool access for multi-step tasks like deep code reviews, test generation, or documentation writing. Subagents prevent context pollution and ensure each task runs with its own focused scope and configuration.

Yes, Subagents excel at documentation writing and data analysis by isolating these tasks with their own system prompts and tool selections. Define the task scope, configure appropriate tools, and delegate—your Subagent executes independently with defined responsibilities and output formats.

Choose based on task complexity: Haiku for lightweight, focused tasks; Sonnet for balanced performance on multi-step work; Opus for complex analysis requiring deeper reasoning. This Skill guides model selection aligned with your Subagent's specific responsibilities and performance requirements.