Trait SgrAgent 

pub trait SgrAgent {
    type Action;
    type Msg: AgentMessage;
    type Error: Display;

    // Required methods
    fn decide(
        &self,
        messages: &[Self::Msg],
    ) -> impl Future<Output = Result<StepDecision<Self::Action>, Self::Error>> + Send;
    fn execute(
        &self,
        action: &Self::Action,
    ) -> impl Future<Output = Result<ActionResult, Self::Error>> + Send;
    fn action_signature(action: &Self::Action) -> String;
}

Base SGR Agent trait — implement per project.

Covers the non-streaming case (va-agent, simple CLIs). For streaming, implement SgrAgentStream on top.

Stateful executors

If execute needs mutable state (e.g. MCP connections), use interior mutability (Mutex, RwLock). The trait takes &self to allow concurrent action execution in the future.

Required Associated Types

type Action

The action union type (BAML-generated, project-specific).

type Msg: AgentMessage

The message type (implements AgentMessage).

type Error: Display

Error type.

Required Methods

fn decide( &self, messages: &[Self::Msg], ) -> impl Future<Output = Result<StepDecision<Self::Action>, Self::Error>> + Send

Call LLM to decide next actions.

fn execute( &self, action: &Self::Action, ) -> impl Future<Output = Result<ActionResult, Self::Error>> + Send

Execute a single action. Returns tool output + done flag. Does NOT push to session — the loop handles that.

fn action_signature(action: &Self::Action) -> String

String signature for loop detection.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety", so this trait is not object safe.

Implementors