agent-expenditure

This skill addresses the issue of uncontrolled token consumption in multi-agent parallel scheduling. In real-world development, when multiple AI agents are dispatched simultaneously to perform tasks, coordination overhead and redundant work lead to token waste, which is often difficult for developers to quantify. This tool tracks each agent's token usage, incorporates Brooks's Law analysis, and automatically identifies and flags waste signals.

Usage is straightforward: simply run this skill after completing parallel agent scheduling. It automatically reads each agent's token consumption data and generates a report containing information such as the number of agents, coordination overhead percentage, and waste signals. You can then query specific agents' token usage details via natural language instructions or obtain recommendations for reducing the number of agents.

It is well-suited for teams or individuals who frequently perform multi-agent parallel scheduling—especially those already experiencing token cost overruns, redundant work between agents, or uncertainty about the optimal number of agents. Agent developers and operations engineers can use it to optimize resource allocation, while project managers can make more informed scheduling decisions based on data.

We recommend running this tool after each parallel scheduling session to establish a cost-monitoring habit. Note that it is primarily designed for post-hoc analysis and is not suitable for real-time monitoring; it also requires the Claude Code environment. For single-agent workflows or scenarios requiring real-time token budget control, we advise using it alongside other tools.