Problem
Persistent goal continuation can incorrectly terminate when the assistant's final response explicitly says the objective is not complete, has remaining open items, or requires continued work.
Expected behavior
If a response contains an explicit incompletion signal, the goal runner should keep the goal active (subject to max-turn/budget limits) rather than marking the goal achieved.
Current downstream workaround
A local downstream patch adds a guard that detects explicit incompletion phrasing before accepting the goal as complete, plus regression coverage for the goal runner.
Notes
This is separate from general judge quality: it covers the concrete case where the response itself says the task is unfinished.
Problem
Persistent goal continuation can incorrectly terminate when the assistant's final response explicitly says the objective is not complete, has remaining open items, or requires continued work.
Expected behavior
If a response contains an explicit incompletion signal, the goal runner should keep the goal active (subject to max-turn/budget limits) rather than marking the goal achieved.
Current downstream workaround
A local downstream patch adds a guard that detects explicit incompletion phrasing before accepting the goal as complete, plus regression coverage for the goal runner.
Notes
This is separate from general judge quality: it covers the concrete case where the response itself says the task is unfinished.