create-or-audit-subagent

You can audit existing subagents by running comprehensive quality gates that check specificity, overlap with built-in agents, and tool scoping. This identifies redundant or misconfigured agents in your .claude/agents directory and generates a report recommending which agents to keep, revise, or remove.

Build codebase-specific subagents by validating isolation needs, investigating project patterns, and writing system prompts that teach real file paths and conventions. Apply configuration hygiene for model selection, tool scoping, and memory settings to ensure agents remain focused and isolated for heavy-output tasks.

You need isolated subagents for heavy-output tasks like code review, test execution, and documentation analysis when engineers require restricted tools and focused personas. Validating against specificity thresholds ensures the subagent adds value beyond Claude's built-in capabilities for your specific project.

The best way to audit subagent configuration hygiene is to run validation checks for model selection, tool scoping, and memory settings. This quality gate includes specificity tests and overlap checks with built-in agents to identify misconfigured agents that lack proper tool access restrictions.

Yes, subagent validation runs overlap checks with built-in agents to identify custom agents that duplicate Claude's default capabilities. This eliminates the risk of cluttering the .claude/agents directory with redundant subagents that add no project-specific value.