What is the passing threshold for an evaluation?

A score of 3.5/5.0 is generally considered the threshold for acceptable quality, while critical operations often require a 4.0 or higher.

Does the Judge skill automatically change my code?

No, this is a report-only command. It provides a detailed evaluation and actionable feedback but does not modify your files directly.

Can I customize what the judge evaluates?

Yes, you can provide an argument-hint to focus the judge on specific areas like security, performance, or accessibility.

How does context isolation work in this skill?

The Judge sub-agent is provided only with the specific task, output, and relevant files, rather than the entire conversation history, which prevents it from being biased by previous interactions.

Judge Sub-Agent

Name: Judge Sub-Agent
Author: NeoLabHQ

byNeoLabHQ

•

542

•

Security & Testing

Evaluates and scores AI-generated work using a specialized, context-isolated sub-agent to ensure high-quality results.

The Judge skill implements the LLM-as-Judge pattern to provide objective, structured evaluations of work produced during a development session. By spawning a sub-agent with isolated context, it eliminates confirmation bias and applies a weighted, multi-dimensional rubric to assess instruction following, solution quality, and reasoning. Each evaluation is evidence-based, requiring specific file citations and justifications before a score is assigned, making it an essential tool for auditing complex code changes, documentation, or architectural decisions before they are finalized.

Key Features

01Evidence-Based Citations requiring specific file and line number references

02542 GitHub stars

03Chain-of-Thought Scoring for 15-25% improved reliability

04Automated Score Interpretation with actionable improvement recommendations

05Multi-Dimensional Rubrics covering code quality, security, and logic

06Context Isolation to prevent session-state confirmation bias

Use Cases

01Auditing complex refactors to ensure all original requirements were met

02Reviewing documentation and technical guides for clarity and completeness

03Evaluating the security and performance of newly generated code snippets

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add neolabhq/context-engineering-kit judge

For use in Claude.ai and ChatGPT

Download Skill