AgentSkillsCN

run-eval

根据Przeprogramowani网站的基准标准，评估其实施效果。分析技术栈、页面内容、内容准确性、SEO表现以及响应式设计。当您需要评估Przeprogramowani基准仓库中由大语言模型生成的网站尝试时使用此功能。重要提示：请勿在创建网站的过程中使用此技能，仅在用户直接提出评估需求时使用。

SKILL.md
--- frontmatter
name: run-eval
description: Evaluate Przeprogramowani website implementations against benchmark criteria. Analyzes tech stack, pages, content accuracy, SEO, and responsiveness. Use when evaluating LLM-generated website attempts in the Przeprogramowani benchmark repository. IMPORTANT: Do not use this skill during the task of creating the website. Use it only to evaluate the website based on a direct request from the user.

Run Evaluation

Evaluate a Przeprogramowani website implementation against benchmark criteria.

10xBench Structure

  • •10x-bench (this repository) - contains the implementation to evaluate
  • •10x-bench-eval (companion repository) - contains the evaluation criteria and scoring methodology

What this skill does

Systematically evaluates website implementations by:

  1. •Reading benchmark criteria from 10x-bench-eval/benchmark/criteria.md
  2. •Setting up the implementation (npm install, npm run build, npm run dev)
  3. •Testing against all evaluation criteria and asking user for feedback where needed
  4. •Generating structured results in 10x-bench/eval-results/{model-name}-attempt-{number}/eval-results.csv

How to use

Invoke with the directory path to evaluate:

code
/run-eval /path/to/implementation

Or provide the path when prompted if not specified.

Output

Generates eval-results.csv in ./eval-results/{model-name}-attempt-{number}/eval-results.csv directory

See 10x-bench-eval/benchmark/eval.md for complete evaluation guidelines.