Struct AgentRunner 

pub struct AgentRunner { /* private fields */ }

Individual agent runner for executing specialized agent tasks

Implementations

impl AgentRunner

pub async fn execute_task( &mut self, task: &Task, contexts: &[ContextItem], ) -> Result<TaskResults>

Execute a task with this agent

impl AgentRunner

pub async fn execute_task_with_retry( &mut self, task: &Task, contexts: &[ContextItem], max_retries: u32, ) -> VtCodeResult<TaskResults>

Execute a task with automatic retry on transient failures

Wraps execute_task with retry logic using exponential backoff. Retries only occur for transient errors (timeouts, network issues, 5xx errors). Non-retryable errors (auth failures, invalid requests) fail immediately.

impl AgentRunner

pub async fn new( agent_type: AgentType, model: ModelId, api_key: String, workspace: PathBuf, session_id: String, settings: RunnerSettings, steering_receiver: Option<UnboundedReceiver<SteeringMessage>>, ) -> Result<Self>

Create a new agent runner.

pub async fn new_with_bootstrap( agent_type: AgentType, model: ModelId, api_key: String, workspace: PathBuf, session_id: String, settings: RunnerSettings, steering_receiver: Option<UnboundedReceiver<SteeringMessage>>, bootstrap: ThreadBootstrap, vt_cfg: Option<VTCodeConfig>, openai_chatgpt_auth: Option<OpenAIChatGptAuthHandle>, ) -> Result<Self>

Create an agent runner with prebuilt thread bootstrap, config, and auth.

pub fn set_quiet(&mut self, quiet: bool)

Enable or disable console output for this runner.

pub fn set_subagent_mode(&self, is_subagent: bool)

Configure the loop detector for subagent mode with tighter read-only budgets and earlier navigation streak intervention.

pub fn set_subagent_controller(&self, controller: Arc<SubagentController>)

Attach a subagent controller to this runner’s tool registry so the subagent-lifecycle tools (agent and its aliases) become available to the model. Used to enable nested delegation: the attached controller carries the child-scoped depth so the depth check in spawn_with_spec still governs grandchild spawns.

pub fn session_messages(&self) -> Vec<Message>

Snapshot the runner-owned conversation messages for archive persistence.

pub fn thread_handle(&self) -> ThreadRuntimeHandle

Clone the underlying thread handle so callers can capture snapshots.

pub fn workspace(&self) -> &Path

Workspace root this runner operates within.

pub fn enable_planning(&self)

Enable read-only planning workflow for the underlying tool registry.

pub fn disable_planning(&self)

Disable read-only planning workflow for the underlying tool registry.

pub fn set_event_handler<F>(&mut self, handler: F)

where F: FnMut(&ThreadEvent) + Send + 'static,

Attach a callback that will be invoked for each structured event as it is recorded.

pub fn clear_event_handler(&mut self)

Remove any previously registered structured event callback.

pub fn set_system_prompt(&mut self, system_prompt: impl Into<String>)

Override the composed system prompt for downstream embedders.

Recomputes system_prompt_report against the overridden text so the budget warning and preflight check stay accurate even when the prompt bypasses the normal section-based composition pipeline.

pub fn tool_registry(&self) -> ToolRegistry

Clone the underlying tool registry so embedders can register custom tools.

pub fn set_tool_definitions_override( &mut self, definitions: Vec<ToolDefinition>, )

Override the tool definitions used for LLM requests instead of the default registry projection.

pub fn clear_tool_definitions_override(&mut self)

Clear any previously set tool definition override, restoring the default registry projection.

pub fn set_tool_arg_transform( &mut self, transform: Arc<dyn Fn(&str, Value) -> Value + Send + Sync>, )

Set an argument transformer applied to tool calls before validation and execution.

pub fn restrict_to_local_tools(&mut self)

Restrict skill child execution using live trusted registration metadata.

pub fn clear_tool_arg_transform(&mut self)

Clear any previously set tool argument transformer.

pub fn set_active_primary_agent( &mut self, active_primary_agent: ActivePrimaryAgent, )

Apply active primary-agent tool and permission policy to this runner.

pub async fn enable_full_auto(&mut self, allowed_tools: &[String])

Enable full-auto execution with the provided allow-list.

pub async fn review_tool_allowlist( &self, allowed_tools: &[String], ) -> Vec<String>

Restrict an allow-list to tools suitable for strict review-only runs.