Struct AgentHarness 

pub struct AgentHarness { /* private fields */ }

Runs a real agent — real AgentFactory, real session loop, real tool gating and answer validation — against a scriptable adapter.

It is a thin wrapper over the production path: for_agent hands back the same AgentBuilder production uses, so there is no parallel session logic that could drift from what ships. That is the property worth protecting; everything else here is convenience.

let harness = AgentHarness::new(triage_agent);
harness.adapter().call_tool(ANSWER_TOOL_NAME, json!({ "urgency": "high" }));

let triage: Triage = harness
    .for_agent()
    .for_mindset()               // the delegation path, gating and all
    .allow_tools(["read_order"])
    .session()
    .await
    .ask("How urgent is this?")
    .await?;

Implementations

impl AgentHarness

pub fn new(agent: Arc<dyn Agent>) -> Self

pub fn builder(agent: Arc<dyn Agent>) -> AgentHarnessBuilder

pub fn adapter(&self) -> &Arc<MockChatAdapter>

pub fn container(&self) -> &Container

The container the agent’s tools resolve from — also where the agent-tools provider is registered, so a mindset built from it can delegate.

pub fn factory(&self) -> &Arc<AgentFactory>

pub fn for_agent(&self) -> AgentBuilder

The production builder for this agent: chain for_mindset(), allow_tools(), deny_tools(), with_budget(), with_context(), then session().

pub async fn session(&self) -> AgentSession

Shortcut for a session with default gating and budget.