agentic-eval

An evaluator-optimizer pipeline separates generation, evaluation, and optimization to refine AI agent outputs. This Skill automates that pattern by applying structured critique loops and rubric-based assessment to iteratively improve generated content.

A self-critique loop is an automated process where an agent repeatedly critiques its own work to identify and address gaps. For code generation, it enables test-driven, automated refinement to improve code quality and correctness.

Yes, rubric-based evaluation uses scoring rubrics to guide improvements and ensure quality. This Skill applies those rubrics within an automated self-evaluation cycle to produce structured evaluation results.

Yes, structured outputs in JSON format are required for evaluation results. This ensures the automated iterative evaluation and refinement process remains auditable and safe across agents, reports, and analysis.

The best way to automate test-driven code refinement is using an evaluator-optimizer pattern that separates generation from evaluation. This Skill provides code-specific refinement through structured self-evaluation loops.

You should avoid automated self-evaluation when your agent outputs cannot be structured into JSON format. The iterative refinement process requires structured evaluation results to enable safe and auditable improvement cycles.