judge-with-debate

Evaluate solutions through multi-round debate between independent judges until consensus.

1.5k|154|Updated Nov 13, 2025
One-click install
npx skills add https://github.com/NeoLabHQ/context-engineering-kit --skill judge-with-debate
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: judge-with-debate
Source: https://github.com/NeoLabHQ/context-engineering-kit/tree/main/plugins/sadd/skills/judge-with-debate
Command: npx skills add https://github.com/NeoLabHQ/context-engineering-kit --skill judge-with-debate

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Single-pass evaluations of code or documents suffer from individual bias and shallow analysis. This Skill runs a structured multi-agent debate where independent judges challenge each other's assessments until they converge on accurate, evidence-based scores.

Core Features & Use Cases

  • Meta-Judge Specification: A meta-judge generates tailored evaluation rubrics, checklists, and scoring criteria once, shared by all judges across all rounds.
  • Independent Parallel Analysis: Three judges independently score the solution against the specification, preventing groupthink.
  • Iterative Debate Rounds: Up to 3 debate rounds where judges defend positions with evidence, challenge disagreements, and revise scores until consensus.
  • Use Case: After implementing a REST API, run the debate evaluation to get consensus scores on correctness, design, security, performance, and documentation, with a final pass/fail recommendation.

Quick Start

Ask the agent to evaluate your solution file with judge-with-debate, providing the solution path and the task it was supposed to accomplish.

Frequently Asked Questions about judge-with-debate

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

FAQPage Schema
How do I evaluate code quality with multiple AI judges?

Run the judge-with-debate command with your solution path and task description. It dispatches a meta-judge to build evaluation criteria, then three independent judges score the solution and debate disagreements until reaching consensus.

What is multi-agent debate for solution evaluation?

Multi-agent debate is an evaluation pattern where independent judges analyze a solution, then iteratively challenge each other's scores with evidence across up to 3 rounds. This reduces individual bias and produces more accurate assessments than single-pass review.

How does the debate evaluation reach consensus?

Consensus is achieved when all judges' overall scores fall within 0.5 points and no criterion differs by more than 1 point. If consensus fails after 3 debate rounds, the persistent disagreements are flagged for human review.

Where are judge evaluation reports stored?

Reports are written to the .specs/reports/ directory using the naming convention {solution-name}-{date}.{judge-number}.md. Judges append debate round sections to their own report files and read other judges' reports directly from the filesystem.

When should I not use debate-based evaluation?

Avoid it for trivial checks where a single review suffices, since it spawns multiple Opus-model agents across several rounds. It is designed for high-stakes evaluations where rigorous, evidence-based scoring justifies the extra compute.