Internet Draft Specification for Agent Directory Service

[muscariello]

Internet Draft Specification for Agent Directory Service

A new Internet Draft specification has been published that defines infrastructure for discovering and managing AI agents in multi-agent systems:

Draft Specification

Agent Directory Service (ADS)

Document: draft-agntcy-ads
Scope: Distributed directory service for AI agent discovery and metadata management

This document specifies the Agent Directory Service, a distributed directory infrastructure designed for the agentic AI ecosystem. ADS addresses the fundamental challenge of discovering, evaluating, and composing AI agents in large-scale multi-agent systems.

Technical Overview

Core Functionality

ADS enables agent discovery through structured metadata storage and retrieval:

• Capability-based discovery using hierarchical skill taxonomies
• Performance evaluation including cost, latency, and resource requirements
• Multi-agent workflow composition through agent linking and dependency tracking
• Verifiable claims with cryptographic integrity and provenance tracking
• Version management for agent evolution and compatibility

Architecture Components

Storage Layer:

• Content-addressed storage using globally unique identifiers
• ORAS (OCI Registry as Storage) integration for standards compliance
• Multi-registry federation supporting organizational boundaries
• Artifact organization with cross-referencing capabilities

Discovery Layer:

• Distributed Hash Table (DHT) implementation using Kademlia protocol
• Two-level mapping: Skills-to-CID and CID-to-PeerID
• Hierarchical skill taxonomies for efficient search space partitioning
• Multi-dimensional search across skills, domains, and features

Security Model:

• Sigstore integration for keyless signing and content provenance
• OpenID Connect authentication with identity-based verification
• Cryptographic integrity through automatic hash computation
• Content verification and tamper detection

Taxonomy System

ADS employs multiple parallel taxonomies for comprehensive agent classification:

Skills Taxonomy: Functional capabilities (natural language processing, computer vision, analytical skills)
Domain Taxonomy: Application areas (networking, security, healthcare, finance)
Feature Taxonomy: Integration patterns (MCP, Agent-to-Agent communication, evaluation frameworks)

All searches require skill-based criteria as the mandatory foundation, with domain and feature taxonomies providing additional filtering dimensions.

Technical Motivation

Multi-agent systems require specialized infrastructure that addresses:

Discovery Challenges:

• Efficient capability-based search across large agent populations
• Semantic relationship handling between related agent capabilities
• Dynamic agent composition and workflow creation

Scale Requirements:

• Distributed architecture supporting global agent networks
• Content deduplication and efficient storage utilization
• Scalable indexing through distributed hash tables

Security Requirements:

• Content integrity and authenticity verification
• Decentralized trust without long-lived key management
• Audit trails for agent publication and modification history

Integration Framework

ADS leverages the Open Agentic Schema Framework (OASF) for structured agent metadata, enabling queries such as:

• "Find natural language processing agents specialized for healthcare applications"
• "Locate computer vision agents with observability features for manufacturing"
• "Discover agents capable of mathematical reasoning with MCP integration"

The system supports compatibility with existing container ecosystems through OCI standards integration and provides interoperability with mature registry tooling.

Intended Audience

This specification is relevant for:

• Multi-agent system architects and developers
• AI framework implementers (LangGraph, AutoGen, LlamaIndex)
• Enterprise AI platform teams
• Distributed systems researchers
• Standards organizations focused on AI interoperability

Document Access

Specification: https://spec.dir.agntcy.org/draft-agntcy-ads.html

Resources:

• GitHub Repository: agntcy/dir-spec
• Specification Site: https://spec.dir.agntcy.org

Community Participation

This draft specification represents initial work toward standardizing agent discovery infrastructure for multi-agent AI systems. Community input is requested for:

• Technical review and implementation feedback
• Use case validation and deployment scenarios
• Integration experience with existing AI frameworks
• Standards development participation

Comments and feedback can be posted in the GitHub repository: https://github.com/agntcy/dir-spec