Struct ResolvedAgentExecution 

pub struct ResolvedAgentExecution<'a> {

    pub model_access: ResolvedModelAccess<'a>,
    pub tools_registry: &'a [Box<dyn Tool>],
    pub observer: &'a dyn Observer,
    pub silent: bool,
    pub approval: Option<&'a ApprovalManager>,
    pub multimodal_config: &'a MultimodalConfig,
    pub config: Option<&'a Config>,
    pub max_tool_iterations: usize,
    pub hooks: Option<&'a HookRunner>,
    pub excluded_tools: &'a [String],
    pub dedup_exempt_tools: &'a [String],
    pub activated_tools: Option<&'a Arc<Mutex<ActivatedToolSet>>>,
    pub model_switch_callback: Option<ModelSwitchCallback>,
    pub pacing: &'a PacingConfig,
    pub strict_tool_parsing: bool,
    pub parallel_tools: bool,
    pub max_tool_result_chars: usize,
    pub context_token_budget: usize,
    pub receipt_generator: Option<&'a ReceiptGenerator>,
    pub knobs: &'a LoopKnobs,
}

Fields

model_access: ResolvedModelAccess<'a>

Provider + model + temperature.

tools_registry: &'a [Box<dyn Tool>]

The tools available this turn (gated per the agent’s policy upstream).

observer: &'a dyn Observer

Telemetry/audit sink.

silent: bool

Suppress stderr output (subagents/reviews run silent).

approval: Option<&'a ApprovalManager>

Approval policy + back-channel; None for paths that never prompt.

multimodal_config: &'a MultimodalConfig

Vision-model routing config.

config: Option<&'a Config>

Full config, for resolving the configured vision_model_provider’s alias-specific runtime options (the vision override, endpoint URI, credentials) on the vision route. None on configless (test) paths, where the route falls back to the legacy factory.

max_tool_iterations: usize

Agentic loop iteration cap.

hooks: Option<&'a HookRunner>

Lifecycle hooks; None when unconfigured.

excluded_tools: &'a [String]

Tools the policy denies (never invoked).

dedup_exempt_tools: &'a [String]

Tools exempt from call-dedup.

activated_tools: Option<&'a Arc<Mutex<ActivatedToolSet>>>

Activation set for on-demand (tool_search) MCP tools; shared so activated tools persist across iterations.

model_switch_callback: Option<ModelSwitchCallback>

Back-channel for the model_switch tool.

pacing: &'a PacingConfig

Loop-detection / ignore-tools / timing policy.

strict_tool_parsing: bool

Reject malformed tool-call protocol.

parallel_tools: bool

Allow concurrent tool execution.

max_tool_result_chars: usize

Truncation limit for tool outputs.

context_token_budget: usize

History-pruning token threshold.

receipt_generator: Option<&'a ReceiptGenerator>

Tool-receipt tracer; None when receipts are off.

knobs: &'a LoopKnobs

Fine-grained loop behavior flags.

Implementations

impl<'a> ResolvedAgentExecution<'a>

pub fn resolve( model_access: ResolvedModelAccess<'a>, io: ResolvedIo<'a>, runtime: ResolvedRuntimeKnobs<'a>, ) -> Self