Describe the bug
When the user manually interrupts a delegated sub-task (via new_task) whose context window is near capacity, and returns to the parent Orchestrator session, all subsequent new_task delegations from that Orchestrator session permanently fail with "Task was interrupted before this tool call could be completed."
To Reproduce
Steps to reproduce the behavior:
1.Start an Orchestrator mode session
2.Successfully delegate several new_task calls to sub-modes — these work normally
3.Delegate a new_task to Debug mode (configured with a high-capacity model like GPT-5.6 Sol High)
4.Debug mode performs extensive investigation, rapidly filling its context window
5.While Debug's context is near capacity, the user manually interrupts the task and switches back to the parent Orchestrator session (to split the work into smaller tasks)
6.From the same Orchestrator session, attempt another new_task delegation to ANY mode
7.The new task immediately returns "Task was interrupted before this tool call could be completed." without the child task ever starting
8.All subsequent new_task calls in that Orchestrator session also fail permanently
Expected behavior
When the user manually stops a child sub-task and returns to the parent Orchestrator, the Orchestrator should be able to delegate new tasks. Each new_task call should create a fresh sub-task context independent of the interrupted session.
Screenshots
N/A
Video
N/A
What version of zoo are you running
3.69.100241
Additional context
Reproduction sequence (exact events from our session):
-Orchestrator (model: mimo-v2.5-pro) successfully delegated ~15 tasks across Code, Debug, Architect, Ask, Project Research, Code-Light modes over ~2-3 hours — all worked correctly.
-Orchestrator delegated a Debug verification task (Debug 1/6). Debug mode was configured with GPT-5.6 Sol High model. Debug performed extensive file reading and analysis, rapidly filling its context window.
The user manually interrupted the Debug task (while context was near full) and switched back to the parent Orchestrator session, intending to split the remaining work into smaller tasks.
-Orchestrator immediately attempted to delegate Debug 2/6 (smaller scope) → "Task was interrupted before this tool call could be completed."
-Orchestrator attempted a trivial Code mode test task (read 5 lines of a file) → same interruption.
-All subsequent new_task calls failed regardless of target mode.
Key observations:
-The failure is NOT mode-specific. After the trigger event, delegating to both Debug and Code failed.
-The Orchestrator's own context is still healthy: read_file, execute_command, write_to_file, attempt_completion all work normally.
-The new_task tool call returns a result (not a tool error), but the child task never executes.
-The trigger is specifically a user-initiated manual interruption of a child task with near-full context, not natural context exhaustion or the number of delegations.
-The child task was using GPT-5.6 Sol High (high-capacity model) which performed extensive investigation, causing rapid context growth.
-Workaround: Start a completely new Orchestrator session. Delegation works normally again.
-Suspected root cause: When the user manually interrupts a child sub-task with near-full context and returns to the parent Orchestrator, the delegation state (session reference, lock, or reopenParentFromDelegation internal state) in the parent session may not be properly cleaned up, causing all subsequent new_task calls to immediately fail.
Describe the bug
When the user manually interrupts a delegated sub-task (via new_task) whose context window is near capacity, and returns to the parent Orchestrator session, all subsequent new_task delegations from that Orchestrator session permanently fail with "Task was interrupted before this tool call could be completed."
To Reproduce
Steps to reproduce the behavior:
1.Start an Orchestrator mode session
2.Successfully delegate several new_task calls to sub-modes — these work normally
3.Delegate a new_task to Debug mode (configured with a high-capacity model like GPT-5.6 Sol High)
4.Debug mode performs extensive investigation, rapidly filling its context window
5.While Debug's context is near capacity, the user manually interrupts the task and switches back to the parent Orchestrator session (to split the work into smaller tasks)
6.From the same Orchestrator session, attempt another new_task delegation to ANY mode
7.The new task immediately returns "Task was interrupted before this tool call could be completed." without the child task ever starting
8.All subsequent new_task calls in that Orchestrator session also fail permanently
Expected behavior
When the user manually stops a child sub-task and returns to the parent Orchestrator, the Orchestrator should be able to delegate new tasks. Each new_task call should create a fresh sub-task context independent of the interrupted session.
Screenshots
N/A
Video
N/A
What version of zoo are you running
3.69.100241
Additional context
Reproduction sequence (exact events from our session):
-Orchestrator (model: mimo-v2.5-pro) successfully delegated ~15 tasks across Code, Debug, Architect, Ask, Project Research, Code-Light modes over ~2-3 hours — all worked correctly.
-Orchestrator delegated a Debug verification task (Debug 1/6). Debug mode was configured with GPT-5.6 Sol High model. Debug performed extensive file reading and analysis, rapidly filling its context window.
The user manually interrupted the Debug task (while context was near full) and switched back to the parent Orchestrator session, intending to split the remaining work into smaller tasks.
-Orchestrator immediately attempted to delegate Debug 2/6 (smaller scope) → "Task was interrupted before this tool call could be completed."
-Orchestrator attempted a trivial Code mode test task (read 5 lines of a file) → same interruption.
-All subsequent new_task calls failed regardless of target mode.
Key observations:
-The failure is NOT mode-specific. After the trigger event, delegating to both Debug and Code failed.
-The Orchestrator's own context is still healthy: read_file, execute_command, write_to_file, attempt_completion all work normally.
-The new_task tool call returns a result (not a tool error), but the child task never executes.
-The trigger is specifically a user-initiated manual interruption of a child task with near-full context, not natural context exhaustion or the number of delegations.
-The child task was using GPT-5.6 Sol High (high-capacity model) which performed extensive investigation, causing rapid context growth.
-Workaround: Start a completely new Orchestrator session. Delegation works normally again.
-Suspected root cause: When the user manually interrupts a child sub-task with near-full context and returns to the parent Orchestrator, the delegation state (session reference, lock, or reopenParentFromDelegation internal state) in the parent session may not be properly cleaned up, causing all subsequent new_task calls to immediately fail.