Discoverable Agent Knowledge -- A Formal Framework for Agentic KG Affordances (Extended Version)

Researchers propose a formal framework for describing knowledge graph affordances to agents, extending decades-old semantic web service standards to address modern KG discovery and composition challenges. The framework introduces the Agentic Affordance Profile (AAP), a metadata layer that enables principled selection and failure diagnosis by specifying what agents can prove from a knowledge graph and under what epistemic conditions.

This academic work addresses a fundamental gap in how knowledge graphs expose their capabilities to autonomous agents. While existing metadata standards like VoID and DCAT describe what data a KG contains, they remain silent on what computational guarantees an agent can rely upon—a critical oversight as KGs increasingly power AI systems. The authors resurrect insights from the 2000s Semantic Web Services movement, when OWL-S and WSMO tackled analogous problems for web service discovery, and adapt these formal methods to contemporary KG environments.

The framework's four dimensions—Semantic Expressivity, Agentic Discoverability, Task-Relative Grounding, and Epistemic Trust Scope—operationalize what agents must know before safely invoking a KG. A key insight is that deployed KGs frequently diverge between their declared schema and actual entailment regime, creating invisible failure modes. The AAP sits above existing metadata standards, enriching them with epistemic guardrails.

For the AI infrastructure sector, this work matters because autonomous agents increasingly compose multiple data sources and services. Without formal affordance descriptions, agents either over-trust KG results or waste resources validating every claim. The framework enables principled matching between agent capabilities and KG guarantees at planning time, reducing both false confidence and computational waste. The scholarly search example demonstrates practical applicability, though the authors acknowledge substantial engineering work remains to operationalize AAP-based matching at scale across heterogeneous KG ecosystems.

• →Current KG metadata standards lack epistemic guarantees about what agents can reliably prove, creating silent failure modes in deployed systems.
• →The proposed Agentic Affordance Profile extends 20-year-old semantic web service standards to modern knowledge graph composition challenges.
• →Four formal dimensions define how agents should evaluate KG trustworthiness: expressivity, discoverability, grounding, and epistemic scope.
• →Framework enables agents to perform principled KG selection and failure diagnosis before attempting task execution.
• →Substantial computational and engineering work remains to scale AAP-based affordance matching across heterogeneous KG ecosystems.