Package dev.agenor.runtime.reflection

Class DefaultReflectionStrategy

java.lang.Object

dev.agenor.runtime.reflection.DefaultReflectionStrategy

All Implemented Interfaces:

ReflectionStrategy

public class DefaultReflectionStrategy
extends Object
implements ReflectionStrategy

Default ReflectionStrategy implementation backed by an LLMProvider.

Builds a critique prompt (custom or DEFAULT_CRITIQUE_PROMPT), calls LLMProvider.chat(dev.agenor.core.llm.LLMRequest) to obtain structured feedback, then parses the quality score from the response text using the pattern score: X.X.

Score parsing rules:

• Pattern score:\s*([0-9]*\.?[0-9]+) is matched case-insensitively.
• Parsed value is clamped to [0.0, 1.0].
• If no score is found the response is treated as requiring revision with a conservative score of 0.5.

Example:

 DefaultReflectionStrategy strategy = new DefaultReflectionStrategy(llmProvider);
 CritiqueResult result = strategy.critique(output, task, ReflectionConfig.defaults()).join();
 

Since:

0.12.0

Field Summary

Fields

Modifier and Type	Field	Description
static final String	DEFAULT_CRITIQUE_PROMPT	Default critique prompt sent to the LLM when no custom prompt is configured.

Constructor Summary

Constructors

Constructor	Description
DefaultReflectionStrategy(LLMProvider llmProvider)	Creates a DefaultReflectionStrategy using the given LLMProvider.

Method Summary

Modifier and Type	Method	Description
CompletableFuture<CritiqueResult>	critique(String originalOutput, String task, ReflectionConfig config)	Critiques originalOutput in the context of task.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

DEFAULT_CRITIQUE_PROMPT

public static final String DEFAULT_CRITIQUE_PROMPT

Default critique prompt sent to the LLM when no custom prompt is configured.

The prompt instructs the model to reply with a score on the last line using the format score: X.X so it can be reliably parsed.

See Also:

• Constant Field Values

Constructor Details

DefaultReflectionStrategy

public DefaultReflectionStrategy(LLMProvider llmProvider)

Creates a DefaultReflectionStrategy using the given LLMProvider.

Parameters:

llmProvider - the provider used to call the critique LLM (non-null)

Method Details

critique

public CompletableFuture<CritiqueResult> critique(String originalOutput,
 String task,
 ReflectionConfig config)

Critiques originalOutput in the context of task.

Builds a critique prompt, calls the LLM, and parses the score from the response. If the score meets or exceeds ReflectionConfig.scoreThreshold() the returned CritiqueResult has shouldRevise = false; otherwise true.

Specified by:

critique in interface ReflectionStrategy

Parameters:

originalOutput - the text produced by the agent that should be evaluated (non-null)

task - the original task or prompt that produced the output (non-null)

config - reflection parameters controlling thresholds and iteration limits (non-null)

Returns:

a future that resolves to a CritiqueResult; never null