You've hit your session limit · resets 12:30pm

Status Open
Reported on v2.1.296
Maintainer reply None cached
Activity 0 comments · opened Oct 14, 2026

Bug Description
Title: Sub-agents abruptly terminate on 5-hour usage limit without state recovery, causing total task and token loss

Summary:
When running multi-agent workflows, reaching the 5-hour usage quota causes all active sub-agents to terminate immediately. Because there is no graceful pause, state persistence, or checkpointing when the limit is hit, all in-progress tasks fail completely. This results in more than 50% of the allocated usage and budget being consumed with zero completed output.

Steps to Reproduce:

  1. Start a long-running script or unattended overnight session utilizing multiple sub-agents.
  2. Allow the sub-agents to run until the 5-hour usage quota threshold is reached (typically occurs after ~20 minutes of concurrent sub-agent execution).
  3. Observe the state of the sub-agents and the parent workflow once the quota limit is triggered.

Expected Behavior:

  • Sub-agents should either pause execution until the quota resets, save their current state/checkpoints, or wind down gracefully so completed sub-tasks are preserved.
  • Usage limits should accommodate longer, unattended batch workflows without destroying in-progress work.

Actual Behavior:

  • All active sub-agents are hard-terminated the moment the 5-hour usage threshold is reached.
  • Unfinished tasks are completely lost, wasting all tokens and compute consumed up to the point of failure.
  • In unattended overnight runs, the cycle of running for ~20 minutes and hitting the 5-hour window results in repeated hard failures and zero completed tasks despite exhausting the usage allowance.

Impact:

  • High token and financial waste: Over half of the session usage is spent on tasks that get killed before completion.
  • Unattended and overnight multi-agent workflows are currently unusable under the strict 5-hour window.

Suggested Resolution:

  1. Implement graceful degradation or pause-and-resume functionality when a session hits the usage cap, rather than killing sub-agent processes outright.
  2. Replace or supplement the 5-hour rolling quota with a longer window (e.g., a 48-hour quota) so developers can run sustained multi-agent workloads without hitting arbitrary short-term ceilings that invalidate in-progress work.

Environment Info

  • Platform: linux
  • Terminal: tmux
  • Version: 2.1.296
  • Feedback ID: 313bdfa9-0a30-4a93-94f5-c43a23c3da85

View original on GitHub ↗