Struct LoopGuard 

pub struct LoopGuard { /* private fields */ }

Circuit breaker that watches for repetitive tool call patterns and enforces a maximum iteration count.

Tracks call fingerprints, outcome fingerprints, and recent call history to detect various pathological patterns including ping-pong cycles.

Implementations

impl LoopGuard

pub fn new(max_iterations: usize, _repetition_threshold: usize) -> Self

Create a new loop guard with default configuration.

pub fn with_config(config: GuardConfig) -> Self

Create a new loop guard with explicit configuration.

pub fn record_tool_calls(&mut self, tool_calls: &[ToolCall]) -> LoopGuardVerdict

Record a set of tool calls and return a verdict.

This is the main entry point for the guard. Call this before executing each batch of tool calls from the LLM.

pub fn evaluate_call(&mut self, tool_call: &ToolCall) -> GuardVerdict

Evaluate a single tool call and return a graduated verdict.

Use this for finer-grained control where you want to handle Block vs CircuitBreak differently.

pub fn record_outcome(&mut self, tool_call: &ToolCall, result: &str)

Record the outcome of a tool call (call hash + result hash).

If the same outcome has been seen before, it is auto-blocked for future iterations.

pub fn is_outcome_blocked(&self, tool_call: &ToolCall, result: &str) -> bool

Check if an outcome would be blocked (same call + same result seen before).

pub fn record_iteration(&mut self) -> LoopGuardVerdict

Record a non-tool-call iteration (e.g. text-only response check).

pub fn iterations(&self) -> usize

Current iteration count.

Trait Implementations

impl Debug for LoopGuard

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.