Struct CodeActAgentBuilder 

pub struct CodeActAgentBuilder { /* private fields */ }

Available on crate features agents and codeact only.

Builder for CodeActAgent.

model and runtime are required; every other setting has a default. The configuration surface mirrors LlmAgentBuilder (instructions, tools, toolsets, retries, guardrails, callbacks, transfer, output schema, …).

Example

use std::sync::Arc;
use std::time::Duration;
use adk_agent::codeact::CodeActAgent;

let agent = CodeActAgent::builder()
    .name("assistant")
    .model(model)
    .runtime(runtime)
    .temperature(0.2)
    .tool_timeout(Duration::from_secs(30))
    .require_tool_confirmation("delete_file")
    .build()?;

Implementations

impl CodeActAgentBuilder

pub fn new() -> CodeActAgentBuilder

Create a builder with default settings.

pub fn name(self, name: impl Into<String>) -> CodeActAgentBuilder

Set the agent name.

pub fn description(self, description: impl Into<String>) -> CodeActAgentBuilder

Set the agent description.

pub fn model(self, model: Arc<dyn Llm>) -> CodeActAgentBuilder

Set the model (required).

pub fn runtime(self, runtime: Arc<dyn CodeRuntime>) -> CodeActAgentBuilder

Set the code runtime (required).

pub fn tool(self, tool: Arc<dyn Tool>) -> CodeActAgentBuilder

Register a tool callable from scripts.

pub fn sub_agent(self, agent: Arc<dyn Agent>) -> CodeActAgentBuilder

Add a sub-agent this agent may transfer control to.

When any sub-agent is present (or the Runner supplies transfer targets), the transfer_to_agent output is described to the model.

pub fn toolset(self, toolset: Arc<dyn Toolset>) -> CodeActAgentBuilder

Add a toolset whose tools are resolved fresh on each invocation from the run’s ReadonlyContext (e.g. per-user or pooled tools). Resolved tools are merged with static tools; duplicate names are rejected at run time.

pub fn with_skills(self, index: SkillIndex) -> CodeActAgentBuilder

Available on crate feature skills only.

Set a preloaded skills index. The most relevant skill for each user query is injected into the system prompt.

pub fn with_auto_skills(self) -> Result<CodeActAgentBuilder, AdkError>

Available on crate feature skills only.

Auto-load skills from .skills/ in the current working directory.

Errors

Returns an error if the skills directory cannot be loaded.

pub fn with_skills_from_root( self, root: impl AsRef<Path>, ) -> Result<CodeActAgentBuilder, AdkError>

Available on crate feature skills only.

Load skills from .skills/ under a custom root directory.

Errors

Returns an error if the skills directory cannot be loaded.

pub fn with_skill_policy(self, policy: SelectionPolicy) -> CodeActAgentBuilder

Available on crate feature skills only.

Customize skill selection behavior.

pub fn with_skill_budget(self, max_chars: usize) -> CodeActAgentBuilder

Available on crate feature skills only.

Limit injected skill content length (default 2000 chars).

pub fn disallow_transfer_to_parent(self, disallow: bool) -> CodeActAgentBuilder

Prevent this agent from transferring control back to its parent.

Applies only to the parent the Runner supplies via RunConfig::parent_agent; sub-agents are unaffected.

pub fn disallow_transfer_to_peers(self, disallow: bool) -> CodeActAgentBuilder

Prevent this agent from transferring control to peer agents.

Applies only to the runner-provided peers in RunConfig::transfer_targets; sub-agents are unaffected.

pub fn tool_confirmation_policy( self, policy: ToolConfirmationPolicy, ) -> CodeActAgentBuilder

Set the tool confirmation policy.

pub fn require_tool_confirmation( self, tool_name: impl Into<String>, ) -> CodeActAgentBuilder

Require human confirmation for a specific tool.

pub fn require_tool_confirmation_for_all(self) -> CodeActAgentBuilder

Require human confirmation for every tool call.

pub fn generate_content_config( self, config: GenerateContentConfig, ) -> CodeActAgentBuilder

Set the generation config (temperature, tokens, etc.) applied to every model request.

pub fn temperature(self, temperature: f32) -> CodeActAgentBuilder

Set the default sampling temperature for model requests.

pub fn top_p(self, top_p: f32) -> CodeActAgentBuilder

Set the default top-p for model requests.

pub fn top_k(self, top_k: i32) -> CodeActAgentBuilder

Set the default top-k for model requests.

pub fn max_output_tokens(self, max_tokens: i32) -> CodeActAgentBuilder

Set the default maximum output tokens for model requests.

pub fn tool_timeout(self, timeout: Duration) -> CodeActAgentBuilder

Set the per-tool execution timeout (default: 5 minutes).

pub fn output_key(self, key: impl Into<String>) -> CodeActAgentBuilder

Set a session state key under which the final result value is stored.

pub fn before_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add a before-agent callback.

Callbacks run in registration order before the loop starts; the first to return Some(content) short-circuits the run with that content.

pub fn after_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add an after-agent callback.

Callbacks run after the loop completes (not on suspension); the first to return Some(content) emits that content and wins.

pub fn before_model_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, LlmRequest) -> Pin<Box<dyn Future<Output = Result<BeforeModelResult, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add a before-model callback, invoked before each model request.

A callback may rewrite the request (BeforeModelResult::Continue) or skip the model call entirely with a synthetic response (BeforeModelResult::Skip).

pub fn after_model_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, LlmResponse) -> Pin<Box<dyn Future<Output = Result<Option<LlmResponse>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add an after-model callback, invoked after each model response is assembled. Returning Some(response) replaces the response the loop uses to extract the next script.

pub fn before_tool_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add a before-tool callback, invoked before each tool call. The first to return Some(content) short-circuits the call; that content is converted to a value and fed back into the script as the tool’s result.

pub fn after_tool_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add an after-tool callback, invoked after each tool attempt resolves. The first to return Some(content) replaces the result fed back to the script.

pub fn after_tool_callback_full( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, Arc<dyn Tool>, Value, Value) -> Pin<Box<dyn Future<Output = Result<Option<Value>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add a rich after-tool callback receiving the tool, arguments, and response value. Returning Some(value) replaces the result fed back to the script. These run after the plain after_tool_callback chain.

pub fn on_tool_error( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, Arc<dyn Tool>, Value, String) -> Pin<Box<dyn Future<Output = Result<Option<Value>, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Add an on-tool-error callback. When a tool ultimately fails, callbacks run in order; the first to return Some(value) supplies a fallback result (fed back as if the tool succeeded). If none do, the error is raised into the script.

pub fn default_retry_budget(self, budget: RetryBudget) -> CodeActAgentBuilder

Set the default retry budget applied to tools without a per-tool override.

pub fn tool_retry_budget( self, tool_name: impl Into<String>, budget: RetryBudget, ) -> CodeActAgentBuilder

Set a per-tool retry budget override (takes precedence over the default).

pub fn circuit_breaker_threshold(self, threshold: u32) -> CodeActAgentBuilder

Set the circuit-breaker threshold: after this many consecutive failures in one invocation, a tool is short-circuited with an immediate error until the next run.

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

Set a JSON schema describing this agent’s input (descriptive metadata, e.g. for exposing the agent as a tool; not enforced at runtime).

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

Require the final result to validate against this JSON schema. On a mismatch the model is asked to correct it, up to output_max_retries times.

pub fn output_type<T>(self) -> CodeActAgentBuilder

where T: JsonSchema,

Require the final result to match T’s JSON schema. Convenience over output_schema using schemars.

pub fn output_max_retries(self, retries: usize) -> CodeActAgentBuilder

Set the maximum number of output-schema correction retries (default 3).

pub fn input_guardrails(self, guardrails: GuardrailSet) -> CodeActAgentBuilder

Set guardrails applied to the incoming user content before the run. A block aborts the run; a transform (e.g. PII redaction) rewrites it. No-op unless the guardrails feature is enabled.

pub fn output_guardrails(self, guardrails: GuardrailSet) -> CodeActAgentBuilder

Set guardrails applied to the final result before it is emitted. No-op unless the guardrails feature is enabled.

pub fn instruction(self, instruction: impl Into<String>) -> CodeActAgentBuilder

Set the agent instruction, placed first in the system prompt. Supports {state.key} template injection from session state.

pub fn instruction_provider( self, provider: Box<dyn Fn(Arc<dyn ReadonlyContext>) -> Pin<Box<dyn Future<Output = Result<String, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Set a dynamic instruction provider evaluated per invocation (takes precedence over the static instruction).

pub fn global_instruction( self, instruction: impl Into<String>, ) -> CodeActAgentBuilder

Set a global instruction placed before the agent instruction. Supports {state.key} template injection.

pub fn global_instruction_provider( self, provider: Box<dyn Fn(Arc<dyn ReadonlyContext>) -> Pin<Box<dyn Future<Output = Result<String, AdkError>> + Send>> + Sync + Send>, ) -> CodeActAgentBuilder

Set a dynamic global instruction provider evaluated per invocation (takes precedence over the static global_instruction).

pub fn include_contents(self, include: IncludeContents) -> CodeActAgentBuilder

Control which conversation history the agent sees (default: full).

pub fn max_iterations(self, max: u32) -> CodeActAgentBuilder

Set the maximum number of model turns.

pub fn max_error_chars(self, max: usize) -> CodeActAgentBuilder

Set the maximum error-message length fed back to the model.

pub fn build(self) -> Result<CodeActAgent, AdkError>

Build the CodeActAgent.

Errors

Returns a config error if the model or runtime was not set.

Trait Implementations

impl Default for CodeActAgentBuilder

fn default() -> CodeActAgentBuilder

Returns the “default value” for a type.