Open Agent Schema Framework (OASF)

What OASF is

The Open Agent Schema Framework is a standardized schema system for defining and managing the capabilities of AI agents ^[1]. Instead of describing an agent in free-form prose, it relies on attribute-based taxonomies: an agent's attributes, capabilities and the relationships between them are expressed as structured attributes drawn from taxonomies ^[1]. The basic unit is the record, the primary data structure for representing collections of information and metadata about agentic AI applications ^[1]. In practice, an agent's description becomes a machine-readable record rather than a document written only for people to read.

Where it comes from

The framework follows the modeling philosophy of the Open Cybersecurity Schema Framework and reuses parts of its implementation ^[1]. That lineage shows in its shape: a schema-first approach in which a shared vocabulary is defined once and then extended, instead of every system inventing its own fields. OASF is the extensible data model that AGNTCY publishes for describing agent attributes, skills and capabilities ^[3].

Stated goals

Three goals are set out for the framework. The first is a common data structure, so that content about agents can be standardized. The second is unique identification of agents, so that they can be discovered. The third is extension capability, so that third-party features can be added on top of the core ^[1]. The three fit together: a shared structure makes records comparable, identifiers make individual agents findable, and an extension mechanism keeps the shared structure from having to cover every use case itself.

Working across protocols

OASF is intended to describe agent attributes, skills and capabilities across A2A, MCP and other protocols ^[3]. Because the description lives in the schema rather than in any single protocol, it can serve as a common layer above them. AGNTCY uses OASF so that any agent can discover and understand the capabilities of others ^[2]. That is the practical purpose of a shared vocabulary: an agent that has never met another one can still read its record and learn what it can do.

Tooling

The project ships development tooling with schema validation and hot reload ^[1]. Validation checks that a record conforms to the schema; hot reload lets a developer change a schema and see the effect without a full restart. The workflow around this tooling is managed with a Taskfile-based workflow and a containerized environment, so that the toolchain runs the same way for everyone who uses it ^[1].

Where it sits

OASF is a description layer, not an agent runtime and not a communication protocol. Its contribution is to give agents, and the tools built around them, one agreed way to write down what an agent is and what it can do, with room to extend when new capabilities appear ^[1][3]. Whether that vocabulary is enough for a given application depends on how well its taxonomies cover the capabilities involved; the framework's answer to gaps is extension rather than a fixed, closed list ^[1].