llm-judge

What This Skill Does

The llm-judge skill provides an automated framework for comparing code implementations across different repositories. It utilizes an LLM-as-judge methodology to objectively evaluate code quality. The process is broken into two distinct phases: Fact Gathering and Judging. In the first phase, dedicated agents analyze individual codebases based on a provided specification, identifying key architectural and functional details. In the second phase, specialized judge agents evaluate these findings across five core dimensions: functionality, security, test quality, overengineering, and dead code. By using weighted rubrics, the skill aggregates these inputs into a final, actionable score, enabling developers to make data-driven decisions when choosing between different code patterns or library implementations.

Installation

You can install this skill directly via the OpenClaw CLI using the following command:

clawhub install openclaw/skills/skills/anderskev/llm-judge

Ensure that your environment has sufficient permissions to read repository files and write to the .beagle directory.

Use Cases

• Library Selection: Compare multiple open-source implementations to determine which is most secure or best engineered for your needs.
• Code Review Standardization: Use the judge as a baseline for peer reviews to ensure consistent evaluation metrics across a team.
• Refactoring Strategy: Evaluate legacy code against a proposed modern replacement to quantify improvements in test coverage and complexity reduction.
• Benchmarking: Score how different team branches handle specific functional requirements against a shared technical specification.

Example Prompts

1. "/beagle:llm-judge compare ./projects/auth-v1 and ./projects/auth-v2 using the spec at ./specs/auth-reqs.md"
2. "/beagle:llm-judge evaluate repository-a, repository-b, and repository-c against the requirements in ./documentation/api-standard.md"
3. "/beagle:llm-judge rank implementations for the new data pipeline using /specs/pipeline-design.md"

Tips & Limitations

• Precision: The quality of the evaluation is highly dependent on the clarity and detail of your specification document. Be as explicit as possible.
• Compute Costs: Because this skill spawns multiple agents in parallel, ensure you have sufficient token limits configured for your LLM provider.
• Human-in-the-loop: Always treat the generated report as a supportive tool for decision-making rather than the final authority. Review the 'Verdict' section carefully to understand the rationale behind specific score deductions.