7 of My 8 Claude Code Agents Had Zero Calls in 30 Days: Finding Dead Agents Automatically

In a recent setup involving eight custom agents defined in Claude Code, seven of these agents had not been called at all over a period of 30 days. This issue stemmed from Claude Code's ability to define custom agents through .md files stored in the ~/.claude/agents/ directory. Users can create agents like architect, code-reviewer, and security-reviewer, expecting Claude Code to select the appropriate agent automatically.

However, the reality is that unless explicitly specified, Claude Code defaults to a generic "general-purpose" agent or "Explore" option when no other agent is called. In the example provided, of the eight defined agents, only one, code-reviewer, was utilized even once in the last 30 days. The remaining seven agents had zero calls, representing a significant waste of token costs and potentially impacting inference quality.

The underlying issue lies in the fact that Claude Code does not autonomously read the descriptions of unused agents and adhere to their intended instructions. Unused agents not only incur token costs but also contribute to cognitive overhead, as it becomes challenging to manually track which agents are truly functioning within the system.

To address this, the recommended approach is to rely on measured data rather than intuition. By establishing an operational cycle that automatically identifies agents with no usage in a specified period (e.g., 30 days), users can take action to remove or refine these agents. Implementing a script that continuously evaluates agent usage and generates reports on unused agents can help maintain the integrity and efficiency of the automated environment.

This proactive monitoring ensures that resources are allocated effectively, and the system operates with confidence, free from the burden of redundant or non-functional agents.