Struct LoopControlAgent 

pub struct LoopControlAgent { /* private fields */ }

Agent that guards LLM tool-call cycles against runaway iteration.

Insert this agent between a chat agent’s message output and the tool-execution node. It forwards traffic transparently while counting, per flow, the final assistant messages that request tool calls. Once the count would exceed max_iterations, the triggering message is not forwarded — severing the cycle — and a synthesized assistant message explaining the stop is emitted on limit_exceeded instead.

A message is counted only when all of the following hold: role is “assistant”, tool_calls is present and non-empty, streaming is false, and its id differs from the last counted id for the flow. Streaming turns re-deliver the same tool_calls-bearing message under one id (partials plus a possibly duplicated final), so counting every delivery would exhaust the limit within a few turns. Messages without an id cannot be deduplicated and are always counted. Everything else — non-message values, user / tool / system messages, streaming partials, and assistant messages without tool calls — passes through unchanged.

Setting max_iterations to zero or a negative value disables the limit. Counters are kept per flow (ctx_key), capped at 1024 flows with oldest-first eviction, and cleared when the agent stops.

Configuration

• max_iterations - Maximum tool-call iterations per flow; <= 0 disables the limit (default: 25)

Ports

• Input message - Messages flowing through the tool-call cycle
• Output message - The forwarded input while within the limit
• Output limit_exceeded - Synthesized assistant message emitted when the limit is exceeded

Implementations

impl LoopControlAgent

pub const DEF_NAME: &'static str = "modular_agent_core::tool::LoopControlAgent"

pub fn def_name() -> &'static str

pub fn agent_definition() -> AgentDefinition

pub fn register(ma: &ModularAgent)

Trait Implementations

impl AsAgent for LoopControlAgent

fn new( ma: ModularAgent, id: String, spec: AgentSpec, ) -> Result<Self, AgentError>

Constructs a new agent instance.

fn stop<'life0, 'async_trait>( &'life0 mut self, ) -> Pin<Box<dyn Future<Output = Result<(), AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Called when the agent stops.

fn process<'life0, 'async_trait>( &'life0 mut self, ctx: AgentContext, _port: String, value: AgentValue, ) -> Pin<Box<dyn Future<Output = Result<(), AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Processes an input message.

fn configs_changed(&mut self) -> Result<(), AgentError>

Called when configuration values change.

fn start<'life0, 'async_trait>( &'life0 mut self, ) -> Pin<Box<dyn Future<Output = Result<(), AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Called when the agent starts.

impl HasAgentData for LoopControlAgent

fn data(&self) -> &AgentData

fn mut_data(&mut self) -> &mut AgentData