Reflection & Critique Pattern
Agent self-correction loop evaluating tool outputs or generated code against criteria before returning final output.
Mechanism & Definition
An architectural pattern where an agent evaluates its own intermediate outputs before accepting them. A evaluator/critic LLM prompt checks generated code, tool results, or plans against constraint criteria, feeding identified errors back to the generator agent for iterative refinement.