Trait InvocationContext 

pub trait InvocationContext: CallbackContext {
    // Required methods
    fn agent(&self) -> Arc<dyn Agent> ;
    fn memory(&self) -> Option<Arc<dyn Memory>>;
    fn session(&self) -> &dyn Session;
    fn run_config(&self) -> &RunConfig;
    fn end_invocation(&self);
    fn ended(&self) -> bool;

    // Provided methods
    fn is_cancelled(&self) -> bool { ... }
    fn user_scopes(&self) -> Vec<String> { ... }
    fn request_metadata(&self) -> HashMap<String, Value> { ... }
    fn authoritative_transfer_targets(&self) -> bool { ... }
    fn delegation_depth(&self) -> u32 { ... }
    fn max_delegation_depth(&self) -> Option<u32> { ... }
    fn orchestration_root_invocation_id(&self) -> &str { ... }
    fn orchestration_edge_id(&self) -> Option<&str> { ... }
    fn requires_tool_confirmation(&self, _tool_name: &str) -> bool { ... }
    fn get_secret<'life0, 'life1, 'async_trait>(
        &'life0 self,
        _name: &'life1 str,
    ) -> Pin<Box<dyn Future<Output = Result<Option<String>, AdkError>> + Send + 'async_trait>>
       where 'life0: 'async_trait,
             'life1: 'async_trait,
             Self: Sync + 'async_trait { ... }
    fn get_secret_for<'life0, 'life1, 'async_trait>(
        &'life0 self,
        request: &'life1 SecretRequest,
    ) -> Pin<Box<dyn Future<Output = Result<Option<String>, AdkError>> + Send + 'async_trait>>
       where 'life0: 'async_trait,
             'life1: 'async_trait,
             Self: Sync + 'async_trait { ... }
}

Full invocation context available to agents during execution.

Extends CallbackContext with access to the agent itself, memory, session, and run configuration.

Required Methods

fn agent(&self) -> Arc<dyn Agent>

Returns the agent being executed.

fn memory(&self) -> Option<Arc<dyn Memory>>

Returns the memory service, if one is configured.

fn session(&self) -> &dyn Session

Returns the current session.

fn run_config(&self) -> &RunConfig

Returns the run configuration for this invocation.

fn end_invocation(&self)

Signals that this invocation should end after the current turn.

fn ended(&self) -> bool

Returns whether the invocation has been ended.

Provided Methods

fn is_cancelled(&self) -> bool

Returns true if this invocation has been cancelled.

Agents and tools can poll this during long-running work (LLM streaming, HTTP I/O, tool execution) to detect an external cancellation request — for example, a user pressing “Stop” or a call to Runner::interrupt. Checking it at chunk or tool boundaries lets an agent exit promptly and perform any graceful cleanup instead of running to natural completion.

The default returns false. The runtime sets the underlying token when Runner::interrupt() is called or RunConfig::cancellation_token fires.

fn user_scopes(&self) -> Vec<String>

Returns the scopes granted to the current user for this invocation.

When a RequestContext is present (set by the server’s auth middleware bridge), this returns the scopes from that context. The default returns an empty vec (no scopes granted).

fn request_metadata(&self) -> HashMap<String, Value>

Returns the request metadata from the auth middleware bridge, if present.

This provides access to custom key-value pairs extracted from the HTTP request by the RequestContextExtractor.

fn authoritative_transfer_targets(&self) -> bool

Whether the run’s transfer target list replaces static sub-agent targets.

fn delegation_depth(&self) -> u32

Current nested agent-as-tool delegation depth.

fn max_delegation_depth(&self) -> Option<u32>

Maximum nested agent-as-tool delegation depth.

fn orchestration_root_invocation_id(&self) -> &str

Returns the root invocation that owns this orchestration tree.

Nested agent runs use this stable identifier to aggregate budgets, traces, and execution receipts without conflating unrelated runs.

fn orchestration_edge_id(&self) -> Option<&str>

Returns the causal relationship execution that started this run.

fn requires_tool_confirmation(&self, _tool_name: &str) -> bool

Returns whether a runtime-injected tool requires confirmation.

This additive hook lets composition layers protect relationship tools without mutating the concrete agent that receives them.

fn get_secret<'life0, 'life1, 'async_trait>( &'life0 self, _name: &'life1 str, ) -> Pin<Box<dyn Future<Output = Result<Option<String>, AdkError>> + Send + 'async_trait>>

where 'life0: 'async_trait, 'life1: 'async_trait, Self: Sync + 'async_trait,

Retrieve a secret by name from the configured secret provider.

Returns Ok(Some(value)) when a provider is configured and the secret exists, Ok(None) when no provider is configured, or an error on provider failure. The default returns Ok(None).

fn get_secret_for<'life0, 'life1, 'async_trait>( &'life0 self, request: &'life1 SecretRequest, ) -> Pin<Box<dyn Future<Output = Result<Option<String>, AdkError>> + Send + 'async_trait>>

where 'life0: 'async_trait, 'life1: 'async_trait, Self: Sync + 'async_trait,

Resolves a secret for a described access.

A wrapper context must forward this, and a tool context builds the request from the identity the framework gave it. The default drops the description and calls InvocationContext::get_secret, which keeps a context that predates the request object working.

Dyn Compatibility

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors

impl InvocationContext for InvocationContext