Does it provide specific evidence for its scores?

Yes, the skill is required to cite specific file paths, line numbers, or direct quotes from the evaluated text to justify every score it provides.

Can it help improve low-quality AI outputs?

Absolutely. For any output receiving an 'Adequate' or lower verdict, the skill generates a list of 'Required Actions' to guide the agent through necessary improvements.

Is this a replacement for unit tests?

No, this skill focuses on semantic and logical quality. It should be used in tandem with automated testing and linting to provide a complete quality assurance workflow.

How does the weighted scoring work?

The skill calculates a composite score where Accuracy (30%) and Groundedness (25%) are prioritized over Completeness (20%), Coherence (15%), and Helpfulness (10%) to ensure technical reliability.

When is the best time to invoke this skill?

It is best used as a quality gate before marking a task complete, after generating a technical plan, or following a complex code review to ensure high standards.

Agent Output Evaluation & Scoring

Name: Agent Output Evaluation & Scoring
Author: oimiragieo

byoimiragieo

•

Seguridad y Pruebas

Evaluates AI-generated content using a weighted 5-dimension rubric to ensure accuracy, groundedness, and high-quality output.

This skill implements a rigorous LLM-as-judge framework designed to verify the quality of agent outputs before task completion. By scoring content across five critical dimensions—Accuracy, Groundedness, Coherence, Completeness, and Helpfulness—it provides a structured, data-driven approach to AI validation. It replaces subjective assessments with nuanced weighted composite scores and mandatory evidence citations, ensuring that every claim made by an AI agent is verifiable, internally consistent, and actionable within your specific codebase context.

Características Principales

01Weighted composite scoring system for objective quality assessment

025-Dimension Rubric: Accuracy, Groundedness, Coherence, Completeness, and Helpfulness

0319 GitHub stars

04Structured verdict emission with actionable improvement steps for low scores

05Seamless integration with pre-completion quality gates

06Mandatory evidence citations using direct quotes or file references

Casos de Uso

01Validating implementation plans before executing complex code changes

02Ensuring code review comments are grounded in actual codebase evidence

03Performing automated quality checks on documentation and agent reflection cycles

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add oimiragieo/agent-studio agent-evaluation

For use in Claude.ai and ChatGPT

Características Principales

01Weighted composite scoring system for objective quality assessment

025-Dimension Rubric: Accuracy, Groundedness, Coherence, Completeness, and Helpfulness

0319 GitHub stars

04Structured verdict emission with actionable improvement steps for low scores

05Seamless integration with pre-completion quality gates

06Mandatory evidence citations using direct quotes or file references

Casos de Uso

01Validating implementation plans before executing complex code changes

02Ensuring code review comments are grounded in actual codebase evidence

03Performing automated quality checks on documentation and agent reflection cycles

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add oimiragieo/agent-studio agent-evaluation

For use in Claude.ai and ChatGPT