The Six-Role Ecosystem

Agent Format serves a multi-party ecosystem. Understanding who authors, consumes, and enforces each piece of the system is essential for understanding why the schema is shaped the way it is.

Role Definitions

Role	Responsibility	Examples
Agent Owner	Writes .agf.yaml files defining agent identity, interface, action space, constraints, and execution policy selection. Publishes these to a Runtime.	An ML engineer authoring a financial-analyst agent
SDK Developer	Develops and maintains an Agent Format SDK that parses .agf.yaml files and creates runnable agent instances.	Google ADK, LangChain, AutoGen, custom SDKs
Standard Committee	Defines and evolves the Agent Format schema. Develops parser/validator libraries and reference tooling.	The agentformat.org maintainers
Runtime Owner	Operates the runtime that uses an SDK to execute agents and serve Agent User requests. Manages models, tools, credentials, policies, and infrastructure.	A platform team running agent-serving infrastructure
Agent User	Sends requests to the Runtime with an agent ID and input. Receives output. Has no knowledge of .agf.yaml internals.	An end user or upstream service calling the agent API
Governance Team	Authors and manages enterprise security, privacy, and compliance policies. Publishes policies to a registry that the Runtime Owner resolves at execution time. Operates alongside the Runtime Owner with independent authority over policy content.	Enterprise security team, privacy team, cost management team

Interaction Flow

The following diagram illustrates how the six roles interact throughout the agent lifecycle — from authoring an .agf.yaml file through to serving an Agent User request:

Governance Team Detail

The Governance Team operates alongside the Runtime Owner with independent authority over policy content. The Runtime Owner enforces policies but cannot modify or bypass them. This separation ensures that security, privacy, and compliance policies remain intact regardless of runtime operational decisions.

Schema Section Ownership

Each section of the .agf.yaml schema has clear ownership — who writes it, who validates it, who consumes it, and who enforces it:

Schema Section	Authored By	Validated By	Consumed By	Enforced By
schema_version	Agent Owner	Parser library	SDK (compatibility check)	Parser library
metadata	Agent Owner	Parser library	Everyone: discovery, registry, tooling	Parser library
interface	Agent Owner	Parser library	Agent User (contract), SDK (validation)	SDK (schema validation at boundary)
memory	Agent Owner	Parser library	SDK (memory provisioning), Runtime (scope resolution)	SDK + Runtime
constraints	Agent Owner + Governance	Parser library + Runtime	Runtime Owner (enforcement)	SDK + Runtime
action_space	Agent Owner	Parser library	SDK (resolution), Runtime (binds implementations)	Runtime (provides implementations by alias)
execution_policy	Agent Owner (selects) + Runtime Owner (defines)	Parser + Runtime (config schema)	SDK (drives execution loop)	Runtime (executes policy)

To understand the motivation behind this multi-role design, see Why a Standard?. For details on how Agent Format fits alongside other standards, see Comparison with Existing Approaches.