[BUG] subagents seem to slow down the longer they are running.

Resolved 💬 7 comments Opened Jul 27, 2025 by jayleaton Closed Aug 15, 2025

Environment

  • Platform (select one):
  • [x] Anthropic API
  • [ ] AWS Bedrock
  • [ ] Google Vertex AI
  • [ ] Other: <!-- specify -->
  • Claude CLI version: latest
  • Operating System: buntu 22.04
  • Terminal: any / warp / iterm2 / cursor

Bug Description

After a while the subagents get unusably slow. This becomes more frustrating when they do not remember the point they were upto. I have sort of worked around this by creating an orchestration subagent that tells the other subagent what they need to do next but if any of the other subagents have not updated their TODO lists then it defeats the purpose.

I have hit rate limits twice now because of this. I never hit rate limits in the past so I am not a power user at all.

I have 2 terminals running right now 1 has had 3 subagents running for over 2 hours. Maybe every 5 minutes they decide to do something.

<img width="493" height="90" alt="Image" src="https://github.com/user-attachments/assets/fff57588-b6e0-44a4-941c-7b78dbad4a24" />

<img width="495" height="55" alt="Image" src="https://github.com/user-attachments/assets/92ea3d92-3d8b-4fb6-b757-26aeade37155" />

Sometimes it just sits and does nothing for several minutes

<img width="513" height="107" alt="Image" src="https://github.com/user-attachments/assets/2ed2a6a1-9eea-4972-9315-18d7450c0d4b" />

Steps to Reproduce

Create subagents that work together. IE product, design, dev, testing. Have each pass the tasks to the next agent. After sometime they will start to slow down. This seems to slowdown my entire machine while not as bad its noticiable on smaller VM's that I have tesed on

Expected Behavior

SHould just work at the same performance as normal agents

Actual Behavior

Slows to an eventual halt of everything

Additional Context

<!-- Add any other context about the problem here, such as screenshots, logs, etc. -->

View original on GitHub ↗

This issue has 7 comments on GitHub. Read the full discussion on GitHub ↗