​
How does YAFAI-Skills work?

1

Prepare the skill manifest

This step involves creating a skill manifest that defines the tools and third-party integrations available to the agents. The manifest includes details about each tool, such as its name, description, and how it can be invoked.

2

Spin up the skill server

This spins up a unix socket that core can connect to and request available tools and integrations. The skill server listens for requests from the YAFAI-Core and responds with the appropriate tool or integration details.

3

Agent asks for available tools

After the skill server is up, agents can connect to it to access the tools and integrations defined in the manifest. This allows agents to perform actions and retrieve information from external systems.

4

Agent invokes a tool or integration based on LLM recommendation

When an agent needs to perform an action that requires a tool or integration, it sends a request to the skill server. The server processes the request and executes the appropriate tool or integration, returning the result to the agent.

​
How YAFAI-Skills handles the tool or integration execution?

Skill Server is a gRPC service with the following proto,

Skill Server RPCs

service SkillService {
  rpc GetActions (GetActionRequest) returns (GetActionsResponse);
  rpc ExecuteAction (ExecuteActionRequest) returns (ExecuteActionResponse);
}

These two RPCs form the basic interface for fetching, and executing tools for yafai agents. Invoke them via “ANY” gRPC client.

​
GetAction RPC

message GetActionRequest {
  string task = 1;
}

message GetActionsResponse {
  repeated Action actions = 1;
}

​
ExecuteAction RPC

message ExecuteActionRequest {
  string name = 1; // Name of the action to execute
  google.protobuf.Struct queryParams = 2;
  google.protobuf.Struct bodyParams = 3; // Represent body parameters as a map
  google.protobuf.Struct pathParams = 4; // For parameters in the URL path
}

message ExecuteActionResponse {
  string response = 1;
  Value result = 2;
  Error error = 3;
}