critic-judge-design

Design Critic/Judge nodes with input schemas and structured verdicts.

Updated Apr 3, 2026
One-click install
npx skills add https://github.com/Victoriakaey/build-reliable-agents --skill critic-judge-design
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: critic-judge-design
Source: https://github.com/Victoriakaey/build-reliable-agents/tree/main/skills/critic-judge-design
Command: npx skills add https://github.com/Victoriakaey/build-reliable-agents --skill critic-judge-design

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Designing reliable Critic, Judge, or Evaluator nodes for LLM pipelines requires clear input/output contracts and a robust reasoning order to ensure consistent, trustworthy judgments.

Core Features & Use Cases

  • Define a precise primary job for the Critic (what is being judged, the binary outcome, and required inputs).
  • Design the input structure and output schema with reasoning preceding the verdict, and choose between single-pass or multi-stage workflows.
  • Implement deterministic routing, loop guards, and explicit handling of common failure modes to improve reliability.
  • Apply to evaluation tasks such as answer quality assessment, evidence sufficiency checks, or tool-usage validation across stages of an agent.

Quick Start

Provide a clean Critic/Judge design specification for a given evaluation task and output a structured verdict with reasoning before judgment.

Frequently Asked Questions about critic-judge-design

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

FAQPage Schema
How do I design a reliable LLM evaluator node for answer quality assessment?

To design a reliable LLM evaluator, define a precise primary job specifying what is being judged, the binary outcome, and required inputs to ensure consistent, trustworthy judgments across your pipeline.

What is the best output schema for a critic node to ensure reasoning precedes the verdict?

The best output schema for a critic node requires structured reasoning to precede the final verdict, ensuring the LLM evaluates evidence and integrates context before committing to a binary judgment.

How do I implement loop guards and safe defaults in multi-stage LLM feedback loops?

Implement loop guards and safe defaults in multi-stage LLM feedback loops by establishing deterministic routing and explicit handling of common failure modes to improve evaluation reliability.

When do I need a single-pass versus a multi-stage reasoning workflow for LLM evaluation?

You need a single-pass reasoning workflow for direct answer quality assessments, whereas a multi-stage workflow is necessary for complex evidence sufficiency checks and tool-usage validation across agent stages.

Can I use this critic design approach for tool-usage validation across agent stages?

Yes, you can apply this critic design approach to tool-usage validation across agent stages by specifying the input structure and output schema required for deterministic routing and evidence integration.

Why does my LLM judge produce inconsistent judgments across different pipeline runs?

LLM judges produce inconsistent judgments when lacking a defined input structure and output schema, making it necessary to enforce reasoning before the verdict and implement deterministic routing.