World Model, Goal Inference, User Signals Integrated

WorldStateService now uses a MemoryStore cache, refreshed during idle periods and updated by signal handlers, providing ambient context, active conversation topics, and reasoning focus to the reasoning loop.

GoalInferenceService detects recurring topic patterns across conversations, validates them with an LLM, and proposes persistent tasks to the user during idle periods.

User chat messages are now processed as signals routed to reasoning:priority, ensuring user input bypasses background processing to maintain low latency.

The reasoning loop enriches each user message with the current cognitive state before dispatching to digest_worker, and checks the priority queue to abort background reasoning if a user message is pending.

The codebase underwent a major migration, renaming CognitiveDriftEngine to ReasoningLoopService and implementing a signal dispatch table to handle various signal types.

The continuous reasoning loop is now signal-driven, triggered by signals like memory_pressure, new_knowledge, and ambient_context, replacing timer-based polling.

An ONNX skill-selector now replaces LLM skill selection in the cognitive triage pipeline for ACT mode, boosting performance and confidence prediction.

Seven services now emit signals to the reasoning:signals queue, completing Milestone 1 of the continuous reasoning loop migration.

• WorldStateService reads from MemoryStore cache refreshed during idle periods.

• GoalInferenceService detects recurring topic patterns across conversations to create PROPOSED persistent tasks.

• User chat messages flow through the reasoning loop before reaching the response pipeline via reasoning:priority.

• CognitiveDriftEngine was renamed to ReasoningLoopService with a signal dispatch table.

• The reasoning loop is now signal-driven, triggered by memory_pressure, new_knowledge, and ambient_context signals.

• ONNX skill-selector replaces LLM skill[] selection for ACT mode in cognitive triage.