bdi-mental-states_skill

Overview

This skill models agent mental states using a Belief-Desire-Intention (BDI) ontology and patterns to transform RDF context into explainable cognitive representations. It supports bidirectional mapping between RDF triples and internal beliefs, desires, and intentions to enable deliberative reasoning, traceability, and interoperability in multi-agent systems.

How this skill works

The skill ingests external RDF graphs and applies Triples-to-Beliefs translation rules to produce belief, desire, and intention instances, each linked to world-state references and temporal validity. It runs BDI processes (belief formation, desire generation, intention commitment), optionally augments LLM outputs with ontological constraints, and projects resulting mental states back into RDF using Beliefs-to-Triples patterns for downstream consumption. Justifications, parts, and temporal intervals are recorded for explainability.

When to use it

• Converting external RDF context into agent beliefs for reasoning
• Implementing or prototyping BDI architectures (SEMAS, JADE, JADEX)
• Augmenting LLM-driven agents with structured cognitive constraints
• Coordinating mental states across multi-agent platforms
• Tracing decisions with explainable chains from perception to action

Best practices

• Model world states separately and reference them from mental states
• Distinguish persistent mental entities from temporal processes (endurants vs perdurants)
• Attach explicit Justification instances to every mental entity for traceability
• Give every mental state a validity interval for temporal queries
• Use hasPart relations for compositional beliefs to enable selective updates
• Validate RDF output against the ontology before committing projected triples

Example use cases

• A cognitive agent converts incoming event RDF into beliefs, generates desires, and forms intentions with associated plans and tasks
• An LLM pipeline constrained by the BDI ontology to prevent inconsistent or unjustified recommendations
• A multi-agent coordination scenario where beliefs and intentions are shared via FIPA ACL encoded as RDF
• A planner that queries active beliefs at a given timestamp to choose applicable plans
• An explainability tool that traces which perception justified a given intention and which plan it specifies