agentic-eval

Evaluate and refine agent outputs through structured reflection loops.

Updated Apr 8, 2026
One-click install
npx skills add https://github.com/yiwei79/azoth --skill agentic-eval-yiwei79
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agentic-eval
Source: https://github.com/yiwei79/azoth/tree/main/.opencode/skills/agentic-eval
Command: npx skills add https://github.com/yiwei79/azoth --skill agentic-eval-yiwei79

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Agent outputs are often inconsistent and require manual review. This Skill provides structured evaluation patterns and loops to autonomously critique and refine results, improving reliability and alignment with goals.

Core Features & Use Cases

  • Basic Reflection: Agent evaluates and refines its own output through self-critique.
  • Evaluator-Optimizer: Separate generation and evaluation into distinct components to drive iterative improvement.
  • Code-Specific Reflection: Apply test-driven refinement loops for code generation, including running tests and revising code based on failures.
  • Use Case: Governance and quality assurance for AI assistants across complex tasks.

Quick Start

Initiate a reflection loop on a given agent output and iteratively revise it until all criteria are satisfied.

Frequently Asked Questions about agentic-eval

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
What is an agent evaluation and reflection loop?

You can refine agent outputs by applying an evaluator-optimizer pattern, which separates generation and evaluation into distinct components. This drives iterative improvement cycles until all quality criteria and guardrails are satisfied.

How do I apply test-driven refinement loops for code generation?

Yes, structured evaluation integrates with evaluator pipelines using the agentic-eval patterns. It provides structured JSON results, enforcing guardrails and enabling autonomous critique and refinement within existing automation workflows.

Why do my agent outputs remain inconsistent across different runs?

The evaluator-optimizer pattern separates generation and evaluation into distinct components. This separation drives iterative improvement by allowing the evaluator to independently critique the generator's output until criteria are met.

How do I start an iterative reflection loop on an existing agent output?

To start a reflection loop, initiate the process on a given agent output and iteratively revise it. The agent evaluates and refines its own output through self-critique until all specified criteria are satisfied.