AgentSpecs Catalog (CONCEPT:AU-AHE.evaluation.backtest-harness)¶

The AgentSpec Catalog Generator bridges the gap between dynamic agent generation and reproducible, structured deployments.

Overview¶

While the Knowledge Graph natively stores agent templates (TeamConfig, Specialist), enterprise deployments often require portable, JSON-based artifacts. The AgentSpecGenerator compiles OWL-driven topologies into structured JSON files known as AgentSpecs.

Key Benefits¶

Shareable Architectures: Trading bots and research agents can be exported as JSON blueprints and shared across teams.
Semantic Consistency: Every generated AgentSpec is tied to an ontology_class mapping back to this repository's OWL ontologies, ensuring strict typing and validation.
Reproducibility: AgentSpecs guarantee that the same tools, capabilities, and system prompts are assembled identically in every environment.

Implementation Details¶

Located in agent_utilities.core.agentspec_catalog, the generator parses existing nodes in the Knowledge Graph and structures them into the v1.0 AgentSpec JSON schema.

Example¶

from agent_utilities import AgentSpecGenerator

spec = AgentSpecGenerator.generate_spec(
    name="QuantBotAlpha",
    description="Mean-reversion trading bot for equities.",
    tools=["fetch_prices", "execute_trade"]
)

AgentSpecGenerator.export_catalog([spec], "catalog.json")