agent-builder

To design a Claude Code subagent, define a single responsibility with clear boundaries, write a system prompt establishing its role, configure necessary tools like Read or Bash, and select the appropriate model based on task complexity.

Model selection for subagents depends on task complexity: choose Haiku for lightweight operations, Sonnet for standard development tasks, and Opus for complex reasoning or intricate agent design scenarios requiring advanced capabilities.

Best practices for subagent system prompts include clearly defining the agent's role, capabilities, and approaches, writing precise descriptions, maintaining single responsibility, using version control, and continuously iterating on prompt improvements.

Configure tool access for subagents by assigning specific tools like Read, Write, Bash, and Grep based on the agent's defined scope and boundaries, ensuring it only has permissions necessary for its specialized use case.

Yes, building subagents requires familiarity with Claude Code, system prompt engineering, and agent design principles to effectively establish roles, define boundaries, and configure specialized AI assistants.

Subagents fail scope boundaries due to unclear system prompts, excessive tool configuration, or poorly defined responsibilities, requiring iteration on agent design and prompt constraints to enforce single responsibility.