Fixed #

Consciousness Engine

• Fixed workspace admission edge case that could allow duplicate concepts under heavy broadcast load.
• Fixed salience normalization overflow when processing extremely large concept streams.
• Fixed temporal binding cleanup issue causing stale bindings to persist longer than intended.

Added #

Consciousness engine

• Consciousness — main orchestrator built on Global Workspace Theory (Baars, 1988)
• WorkspaceManager — 7±2 concept broadcast buffer with salience-based admission
• AttentionSpotlight — focus/defocus, automatic wander, salience computation
• BindingEngine — temporal and semantic concept binding with co-activation tracking
• EpisodicMemory — event-based autobiographical memory with decay
• SemanticMemory — world-knowledge triple store with confidence scoring
• WorkingMemory — short-term 4±1 buffer with LRU eviction
• MemoryManager — unified retrieval and episodic→semantic consolidation
• SensoryInputProcessor — full perceptual pipeline with modality tagging
• FeatureExtractor — linguistic features: entities, actions, spatial/temporal markers
• PerceptionBuffer — pre-attentive sensory register
• InferenceEngine — forward-chaining, causal, associative, and memory-driven rules
• ConceptualGraph — directed, weighted semantic network with BFS/DFS/spreading activation
• CausalInferenceEngine — Pearl-inspired cause-effect detection from concept streams
• PatternRecognizer — co-occurrence, sequence, and cluster pattern discovery
• CrossModalBinding — multi-sensory fusion across visual/auditory/tactile modalities
• SynchronizationManager — parallel processing stream coordination
• CoherenceManager — workspace consistency scoring
• ConsciousnessPlugin — extensible processing hooks
• ConsciousnessLogger — structured, level-filtered logging
• ConsciousnessMetrics — latency, throughput, and P95 tracking
• ConsciousnessVisualizer — ASCII workspace/graph rendering

Autonomous agent framework (new in v1.0)

• AgentMind + ConsciousnessAgentExtension — upgrade any Consciousness to an autonomous agent via mind.asAgent(provider: ...)
• AgentGoal, AgentTask, AgentDecision, AgentContext, AgentRunResult — typed data model for goal-driven execution
• AgentMemoryStore — inverted word index with composite scoring and LRU eviction
• LLMProvider (abstract) + EchoLLMProvider, MockLLMProvider, HttpLLMProvider — zero-dependency testing and OpenAI-compatible production backends
• LLMCore — prompt assembly, context compression (middle-truncation), JSON parsing, token usage tracking; supports all 5 action types: use_tool, think, complete, replan, error
• Tool, ToolResult, ToolRegistry, ToolRouter — extensible tool system
• BuiltinToolset — 6 ready-to-use tools: search_web, calculate, read_file, write_file, call_api, schedule_task
• _MathParser — AOT-safe recursive-descent evaluator for CalculateTool
• ExecutionDAG — Kahn's topological sort; full task state machine pending → running → succeeded / failed / skipped
• PlanningEngine — LLM-backed JSON decomposition + rule-based fallback; replan() preserves succeeded tasks
• AgentLoopController — full autonomous observe → retrieveMemory → plan → decide → execute → updateMemory → checkComplete cycle
• EnvironmentAdapter (abstract) + NullEnvironmentAdapter, MockEnvironmentAdapter
• AgentLoopEvent broadcast stream with 14 event types
• SelfReflectionModule — 4 detectors: cascade failures, tool loop, stalled progress, thought spiral; optional LLM deep-reflection; history with LRU cap

Examples

• example/basic_consciousness.dart — consciousness quick-start
• example/advanced_awareness.dart — attention control, plugins, metrics
• example/learning_simulation.dart — rule learning and inference
• example/multi_modal_integration.dart — cross-modal binding
• example/autonomous_agent.dart — full agent loop with event stream
• example/multi_tool_agent.dart — all 6 tools + custom tool + self-reflection

Documentation

• doc/AGENT_ARCHITECTURE.md — deep-dive architecture reference
• THEORY.md — scientific foundations (GWT, binding theory, memory models)
• PERFORMANCE_GUIDE.md — tuning tips and benchmarks

Dependencies #

• collection: ^1.17.0
• equatable: ^2.0.5
• http: ^1.2.1
• logging: ^1.2.0
• meta: ^1.9.0
• uuid: ^4.0.0