Struct AgentTool 

pub struct AgentTool { /* private fields */ }

Available on crate feature tools only.

AgentTool wraps an Agent to make it callable as a Tool.

When the parent LLM generates a function call targeting this tool, the framework executes the wrapped agent, captures its final response, and returns it as the tool’s result.

Implementations

impl AgentTool

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

Create a new AgentTool wrapping the given agent.

pub fn with_config(agent: Arc<dyn Agent>, config: AgentToolConfig) -> AgentTool

Create a new AgentTool with custom configuration.

pub fn skip_summarization(self, skip: bool) -> AgentTool

Set whether to skip summarization.

pub fn forward_artifacts(self, forward: bool) -> AgentTool

Set whether to forward artifacts.

pub fn timeout(self, timeout: Duration) -> AgentTool

Set timeout for sub-agent execution.

pub fn input_schema(self, schema: Value) -> AgentTool

Set custom input schema.

pub fn output_schema(self, schema: Value) -> AgentTool

Set custom output schema.

pub fn session_snapshot(self, snapshot: AgentToolSessionSnapshot) -> AgentTool

Choose which parent session data is snapshotted into the isolated child session.

pub fn forward_memory(self, forward: bool) -> AgentTool

Set whether the wrapped agent can access the parent’s memory service.

pub fn forward_shared_state(self, forward: bool) -> AgentTool

Set whether the wrapped agent can access the parent’s parallel shared state.

This defaults to true to preserve the historical AgentTool behavior.

pub fn forward_events(self, forward: bool) -> AgentTool

Set whether child events are emitted through the parent tool context.

The default is false for backward compatibility. State and artifact deltas are merged regardless of this setting.

pub fn failure_mode(self, mode: AgentToolFailureMode) -> AgentTool

Choose whether delegated failures are returned as JSON or propagated.

pub fn propagate_failures(self, propagate: bool) -> AgentTool

Set whether delegated failures should be propagated as tool errors.

pub fn max_delegation_depth(self, max_depth: u32) -> AgentTool

Bound the number of nested agent-as-tool delegations.

pub fn reject_child_handoffs(self, reject: bool) -> AgentTool

Reject child handoff events that cannot be executed by AgentTool itself.

pub fn execute_child_handoffs( self, agents: impl IntoIterator<Item = Arc<dyn Agent>>, ) -> AgentTool

Executes child handoffs using the supplied exact target registry.

Disabled by default for backward compatibility. When enabled, a child transfer remains inside the nested invocation until the terminal member returns, after which its result is returned to the original caller.

pub fn history_max_events(self, max_events: usize) -> AgentTool

Limits the parent history copied into a delegated session.

A value of zero copies no history while still allowing a filtered state snapshot when parent-session forwarding is enabled.

pub fn state_keys( self, keys: impl IntoIterator<Item = impl Into<String>>, ) -> AgentTool

Copies only these exact state keys into a delegated session.

pub fn output_state_keys( self, keys: impl IntoIterator<Item = impl Into<String>>, ) -> AgentTool

Allows delegated state writes only to these exact keys.

pub fn state_merge_policy(self, policy: AgentToolStateMergePolicy) -> AgentTool

Selects how child state writes merge with concurrent parent updates.

pub fn state_merge_exempt_keys( self, keys: impl IntoIterator<Item = impl Into<String>>, ) -> AgentTool

Exempts trusted framework-owned keys from concurrent merge checks.

Output allowlists still apply. This is intended for runtime bookkeeping that is deliberately updated by both the parent and delegated context.

pub fn artifact_prefixes( self, prefixes: impl IntoIterator<Item = impl Into<String>>, ) -> AgentTool

Allows artifact writes only when their names start with one of these prefixes.

Trait Implementations

impl Tool for AgentTool

fn name(&self) -> &str

Returns the unique name of this tool.

fn description(&self) -> &str

Returns a human-readable description of what this tool does.

fn parameters_schema(&self) -> Option<Value>

Returns the JSON Schema for this tool’s parameters, if any.

fn response_schema(&self) -> Option<Value>

Returns the JSON Schema for this tool’s response, if any.

fn is_long_running(&self) -> bool

Indicates whether the tool is a long-running operation. Long-running tools typically return a task ID immediately and complete the operation asynchronously.

fn execute<'life0, 'async_trait>( &'life0 self, ctx: Arc<dyn ToolContext>, args: Value, ) -> Pin<Box<dyn Future<Output = Result<Value, AdkError>> + Send + 'async_trait>>

where 'life0: 'async_trait, AgentTool: 'async_trait,

Executes the tool with the given context and arguments.

fn declaration(&self) -> Value

Returns the tool declaration that should be exposed to model providers.

fn enhanced_description(&self) -> String

Returns an enhanced description that may include additional notes. For long-running tools, this includes a warning not to call the tool again if it has already returned a pending status. Default implementation returns the base description.

fn is_builtin(&self) -> bool

Indicates whether this tool is a built-in server-side tool (e.g., google_search, url_context).

fn required_scopes(&self) -> &[&str]

Returns the scopes required to execute this tool.

fn is_read_only(&self) -> bool

Indicates whether this tool performs no side effects.

fn is_concurrency_safe(&self) -> bool

Indicates whether this tool is safe for concurrent execution.