Struct LlmReflector 

pub struct LlmReflector { /* private fields */ }

A Reflector that asks an LLM to classify tool failures and suggest corrections.

Holds a shared API client handle and an optional system-prompt override. On each call to analyze it builds a single user message describing the failure, calls request_structured::<FailureAnalysis>, and returns the typed analysis.

When the schema_validation feature is enabled and the engine supplies the failing tool’s schema, the reflector validates the model’s Correction::modified_input against that schema and returns ReflectionError::Internal on a mismatch.

Implementations

impl LlmReflector

pub fn new(client: Arc<dyn ApiClient>) -> Self

Construct an LlmReflector backed by the given shared client.

Uses the module’s default system prompt. Override with with_system_prompt.

pub fn with_system_prompt(self, prompt: impl Into<String>) -> Self

Replace the default system prompt with a caller-supplied one.

Builder-style; consumes and returns self. The prompt should still instruct the model to return JSON conforming to the FailureAnalysis schema, since the response is parsed via StructuredOutput.

Trait Implementations

impl Debug for LlmReflector

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

Formats the value using the given formatter.

impl Reflector for LlmReflector

fn analyze( &self, error: &str, tool_name: &str, tool_input: &Value, tool_schema: Option<&ToolSchema>, context: &ReflectionContext, ) -> Pin<Box<dyn Future<Output = Result<FailureAnalysis, ReflectionError>> + Send + '_>>

Analyse a failed tool call.