Monitor Subagents FAQs

Question 1

Why is it important to track compaction events?

Accepted Answer

Compaction occurs when an agent faces context pressure. High compaction counts can lead to degraded output quality, signaling that you should collect results and review the subagent's work carefully.

Question 2

Can I filter the status of specific subagents?

Accepted Answer

Yes, because the monitoring script outputs standard JSON, you can use tools like jq to filter by specific subagent IDs or extract the summary object for quick counts.

Question 3

What is the recommended polling interval for monitoring?

Accepted Answer

To minimize tool call overhead and context impact, it is recommended to use a polling interval of 30-60 seconds rather than checking every few seconds.

Question 4

How are token metrics calculated for subagents?

Accepted Answer

Metrics are extracted from .completion.json files for finished subagents or via session file parsing for active ones. This represents the actual context processed rather than just cumulative API tokens.

Question 5

What does a 'warning' status indicate in subagent monitoring?

Accepted Answer

A warning status indicates that a subagent has reached or exceeded 80,000 tokens (approximately 40% of its context), suggesting the parent agent should consider intervention or collecting partial results.

Monitor Subagents

主な機能

ユースケース

Monitor Subagents

主な機能

ユースケース