You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Handoff instead of compaction: finish the feature (agent source, thread menu, roles, later items) #877
What: long threads (the assistant and the project orchestrators) never compact. When they get large or sit idle, T3 starts a fresh session in the same thread with a short written handover, so the thread keeps its memory and every later call is cheaper.
Why the owner wants it: one long-lived thread per project that the owner talks to; compaction loses detail and large contexts cost usage.
Already built: HO1 setting and pure decision #282, HO2 HandoffService with size and idle triggers #320 (+ #583#605#612#634#666), HO3 the "command" source #775, HO5 seam lines #774, Fork settings rows #544.
Owner decisions already given:#215 rows 2-4 (only long threads; 150K tokens; idle handoff before the prompt cache runs out).
Done when: an orchestrator thread runs for weeks without compaction, each handoff is visible in the thread, and a failed handoff falls back to normal compaction with an alert.
Activity