agent-expenditure

This Skill addresses the challenge of quantifying token waste in multi-agent parallel scheduling. In practical development, when dispatching multiple agents to execute tasks simultaneously, coordination overhead (e.g., file conflicts, duplicated work) can significantly increase token consumption, yet developers often lack a clear view of which agents are inefficient. This tool tracks each agent's token usage, analyzes the proportion of coordination overhead based on Brooks' Law, and automatically flags waste signals.

Usage is straightforward: simply run this Skill after parallel scheduling is complete. It automatically reads agent execution logs, calculates each agent's token consumption and coordination overhead ratio (based on shared file read/write conflicts), and generates a post-hoc review checklist. You can then query specific agent waste details using natural language instructions, such as "Show the three agents with the highest token consumption" or "Check for any duplicated work."

It is particularly suitable for teams or individuals who frequently engage in multi-agent parallel scheduling, especially those experiencing soaring token costs without being able to pinpoint the root cause. For complex projects managing 10 or more agents, this tool can help decide whether to reduce the number of agents to lower coordination overhead. It is also useful during project retrospectives to evaluate the cost-effectiveness of parallel scheduling.

We recommend using this tool for review after each large-scale parallel scheduling session. Note that it is primarily designed for static analysis after parallel scheduling and is not suitable for single-agent workflows or real-time monitoring scenarios. Additionally, the tool relies on the Claude Code plugin environment and requires complete agent execution logs for accurate analysis.