SkyNet Cognitive System

Technical specification of a stateful, personality-developing cognitive assistant designed to break the 'no-memory / no-personality / no-context' limits of traditional assistants. Three temporal layers (sub-500ms conversation pipeline, ~30-min heartbeat, weekly consolidation) coordinate 17 modules: pattern-based cognitive watcher, importance-aware retention with 4-bucket decay, knowledge-graph inference engine, multi-profile 17-trait personality with per-trait decay, dreaming-loop memory synthesis, attention gate with receptivity scoring, social trainer with humor calibration, metacognition with confidence recalibration, prospective memory with Turkish date parsing, surprise→curiosity bridge, topic clustering, mood journal, guidance feedback loop, episodic→semantic gist compression, weekly self-reflection, internal narrative synthesis, and active-learning weak-spot detection — all backed by atomic JSON IO.