agentic-eval

LLM-as-judge self-correction evaluates AI code generation against defined criteria, enabling iterative refinement loops. Agents generate code, test it against specifications, and automatically refine outputs based on test errors to achieve higher quality.

The evaluator-optimizer pattern is an iterative generation mechanism where an AI agent assesses its own outputs against evaluation criteria. It facilitates loops of generation, evaluation, and correction to improve quality-critical tasks like code and reports.

Yes, you can implement reflection patterns for AI evaluation using C# and Microsoft.Extensions.AI. The skill provides specific patterns and techniques for self-evaluation and iterative refinement within the .NET ecosystem.

To iteratively fix C# code errors, an agent generates code for a specification, tests it against provided unit tests, and if it fails, refines the code based on the test errors. This loop continues until quality standards are met.

You need self-reflection for AI-generated reports and analysis when clear evaluation criteria exist and outputs are quality-critical. It allows agents to critically evaluate and iteratively improve their own responses to meet specific standards.