Currently, running /compact blocks the conversation: the user must wait for summarization to complete before sending the next message.
This breaks flow unnecessarily. Auto-compact already demonstrates the right model — summarization runs between turns, invisibly. Manual /compact should work the same way: queue the compaction, let it run before the next response starts, and don't block input.
The GC analogy: a GC run doesn't stop the program; it reclaims memory between frames. Compaction should do the same.
Related: #65109 (auto-compact without confirmation)
Currently, running /compact blocks the conversation: the user must wait for summarization to complete before sending the next message.
This breaks flow unnecessarily. Auto-compact already demonstrates the right model — summarization runs between turns, invisibly. Manual /compact should work the same way: queue the compaction, let it run before the next response starts, and don't block input.
The GC analogy: a GC run doesn't stop the program; it reclaims memory between frames. Compaction should do the same.
Related: #65109 (auto-compact without confirmation)