{"content":"fix: treat Studio assistant-mode sessions as attended for reasoning effort (IDBots @ fix/studio-attended-effort, commit f0f262a3)\n\nRoot cause (2026-11-11 production forensics): isUnattendedAutomationSession classified every non-standard session as unattended, so Studio browser sessions carrying the assistantMode marker never consulted the cerebellum effort selector and every auto turn resolved straight to max — 33/33 Studio turns mapped max while BotWorks stayed off/low on the same models. First visible content p50 174s vs 48s, per-turn thinking ~21K chars vs ~0.5K.\n\nFix: (1) assistant-mode owner chats now take the attended path — per-message cerebellum classification plus the assistant-mode effort rung (off unless P(high)+P(max) escalates); (2) new resolveAssistantModeExpressibleRung pins an assistant-mode 'off' verdict to 'low' on GLM-family routes on BOTH effort chains (primary + fallback-brain resume), because GLM-5.x cannot express 'off' and the responses wire drops the param, leaving server-default thinking in charge; the fallback-route resolveAutoEffortFallback input now carries the fallback model's GLM family too.\n\nTests: coworkUnattendedEffort 7/7 (new classifier-exemption + GLM-floor-pin static checks and behavioral cases), cerebellumEffortSelector 15/15, coworkAssistantModeSessionGate 11/11, turnAckReply+assistantStatusPhrases 44/44, compile:electron + tsc --noEmit clean.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}