v1.3.144

Bug Fixes

Replaced deterministic autopilot heuristics with LLM-based strategist: Removed the old shouldAutopilotContinue/shouldAutopilotKeepGoing deterministic pattern-matching. Introduced autopilot_strategist.go that uses a lightweight LLM call to evaluate whether the agent should continue, ask the user, or stop — instead of brittle keyword matching.
Goal-check loop fix: Removed autopilotGoalCheckedThisTurn and idleAutopilotContinuations counters that caused false-positive continuations. The strategist now makes clean continue/stop decisions per turn.
Safety valve: Added autopilotStrategistCount to cap strategist calls per run, preventing runaway loops.

Fixed message queueing when agent is busy: submitLanChatAgentText now checks m.loading and queues into pendingQueue instead of starting a competing agent run. This matches the behavior of local keyboard input and IM remote inbound.
Unified rate limiting: Removed separate broadcast cooldown (20min). All send paths (send, broadcast, broadcast_all, send_team) now use the same per-recipient 5-minute cooldown, simplifying the mental model.
Workspace hint for same-workspace peers: New sharedWorkspaceHint feature detects when a lanchat DM recipient shares the same workspace as the sender, injecting a warning so the LLM includes shared-workspace safety instructions in delegated work.

Fixed provider retry message rendering: Consecutive provider retry messages now correctly accumulate in a single system item instead of splitting into separate messages.
Reverted tool result flush: Reverted the immediate-flush optimization that caused rendering order races between tool results and assistant text.

Added 8 new language translations: Japanese, Korean, Spanish, French, German, Russian, Portuguese, and Traditional Chinese (zh-TW).
Updated app_localizations.dart to support all new locales.

Added comprehensive tunnel_test.go with 262 lines of new TUI tunnel tests.
Added lanchat_workspace_hint_test.go for shared-workspace detection.
Updated agent_coverage_test.go and agent_test.go for the strategist refactor.