How does this skill help with multi-agent orchestration?

It provides a centralized, script-based way to check the health and progress of all spawned agents simultaneously, ensuring the parent agent knows exactly when work is finished or if an agent has stalled.

What does a 'warning' status indicate for a subagent?

A warning status indicates that a subagent has reached a specific token threshold (typically 80,000 tokens or 40% of context). This suggests the agent is approaching its limit and you should consider intervening or collecting partial results.

How often should I run the monitoring script?

It is recommended to use polling intervals of 30 to 60 seconds. Checking too frequently (e.g., every second) creates unnecessary tool call overhead and can clutter the session history.

What are subagent compaction events?

Compactions occur when an agent hits context pressure and must compress its history. Monitoring these events is crucial because heavy compaction can lead to a degradation in the agent's reasoning quality.

Subagent Status Monitor

Name: Subagent Status Monitor
Author: cowwoc

bycowwoc

•

分析と監視

Tracks active AI subagents, monitoring token usage and context limits to ensure reliable multi-agent orchestration.

The monitor-subagents skill provides essential observability for complex AI workflows involving multiple concurrent sub-processes. It allows parent agents to track the real-time status of spawned subagents, detect when they are approaching context window limits, and identify compaction events that might affect output quality. By providing standardized metrics and status codes through optimized scripts, it enables robust decision-making during long-running tasks, helping you decide when to intervene, decompose work, or collect final results.

主な機能

01Compaction event detection to safeguard output quality

02Context limit and token consumption monitoring

032 GitHub stars

04Integrated token analysis compatible with completion reports

05Real-time subagent status tracking (running, warning, complete)

06Optimized reporting script to minimize parent agent context impact

ユースケース

01Orchestrating complex multi-agent workflows with data-driven status checks

02Determining when to intervene or collect results from subagents

03Monitoring long-running parallel tasks to prevent context overflows

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add cowwoc/claude-code-cat monitor-subagents

For use in Claude.ai and ChatGPT

Download Skill