implement-agent-reflection

To set up a self-correcting loop for LLM-based tasks, use an Evaluator-Generator pattern that critiques initial outputs and iteratively refines them. This structured reflection process prevents subtle flaws in single-shot generations.

Agent reflection is a technique that enhances output quality by forcing critical self-reflection before finalizing results. It uses an evaluator-generator loop to explicitly critique and refine initial LLM outputs iteratively.

To trigger agent reflection on an existing LLM output, use the command `trigger-reflection-on-output <your-output>`. This initiates the evaluator-generator loop to critique and refine the provided content.

You need self-correcting loops for high-fidelity LLM outputs in scenarios like code review or prompt engineering. These tasks require ensuring a high standard of output by preventing subtle flaws through structured reflection.

Yes, the evaluator-generator approach works by taking single-shot LLM outputs and applying an iterative self-critique loop. It refines the initial generation through explicit reflection to achieve high-quality results.