spawn-child-agent

Use this skill when you have a complex, isolated task that would benefit from a dedicated agent loop, or when you need to delegate work to a specialized sub-agent (like a QA bot or Researcher) in a way that allows that sub-agent to further delegate.

Usage

Execute via the Bash tool (run_command). You MUST use the -p (headless) flag.

CRITICAL: Do NOT attempt to spawn an agent using the browser_subagent tool. That tool is for browser automation only and cannot spawn child agents. You MUST use the claude CLI via run_command. If you are about to use browser_subagent to spawn a child agent, STOP and use run_command instead.

claude -p "Your detailed instructions for the child agent here" \
  --allowedTools "Bash,Read,Grep,Glob,find_by_name,view_file,write_to_file,replace_file_content" \
  --model "sonnet" \
  --output-format json

Critical Flags