菜单
Use an LLM as an automated judge to score responses against a rubric. Use when you need nuanced quality evaluation that can't be done with exact match or regex. Triggers include "score with AI", "llm-judge", "rubric-based scoring", "quality evaluation", or any need to rate free-form responses.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/heldernoid/agentic-build-templates --skill llm-judge命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Edit the visual grid layout of a garden bed, assigning crops to cells on a canvas editor. Use when asked to arrange crops in a bed, draw a planting layout, move crops around, visualize companion planting placement, or check for enemy crop adjacency in a specific bed. Triggers include "layout editor", "bed grid", "draw the bed", "assign crops to cells", "companion warning in layout", or similar visual planning tasks.
Plan crop rotations, planting schedules, and companion planting for garden beds and fields. Use when asked to manage a farm, garden, or plot layout; schedule what to plant and when; check companion planting relationships; track rotation history; or generate a printable planting schedule. Triggers include "crop rotation", "planting schedule", "companion planting", "garden bed", "what to plant", "frost dates", or any task involving seasonal crop planning.
Log crop harvests with yield quantity, quality grade, field, and storage destination. Use when asked to record a harvest, check total yield for a crop or field, filter harvest history by date or grade, view analytics charts, or export harvest data. Triggers include "log harvest", "record yield", "harvest entry", "crop yield", "grade breakdown", "field yield", "harvest history", or any task involving tracking what was picked and where it went.
基于 SOC 职业分类
正在显示 SKILL.md
| name | llm-judge |
| description | Use an LLM as an automated judge to score responses against a rubric. Use when you need nuanced quality evaluation that can't be done with exact match or regex. Triggers include "score with AI", "llm-judge", "rubric-based scoring", "quality evaluation", or any need to rate free-form responses. |
Use an LLM model to score responses against a natural language rubric. Returns a 0.0-1.0 score with reasoning.
The judge model receives:
The judge returns a JSON object: {"score": 0.0-1.0, "reason": "brief explanation"}
scoring:
- method: llm-judge
rubric: "The answer clearly explains the return policy and includes the timeframe"
threshold: 0.8 # score >= 0.8 = pass (default: 0.7)
weight: 2 # double weight vs other scorers (default: 1)
endpoint: openai-prod # override default judge endpoint
model: gpt-4o-mini # override default judge model
Write rubrics as positive statements of what a good response contains:
Good rubric:
"The response mentions the 30-day return window and explains the condition requirement"
Bad rubric (vague):
"The response is good"
Good rubric for tone:
"The response is empathetic, acknowledges the customer's frustration, and provides a concrete next step"
Set globally in Settings or environment:
openai-prod (configurable)gpt-4o-mini (configurable - cheaper models work well for judging)Case input is wrapped in XML delimiters before insertion into the judge prompt to prevent injection attacks:
<case_input>
{user's case input}
</case_input>
llm-judge makes one additional LLM call per scored case. For 100 test cases, each with one judge scorer, you incur 100 extra LLM calls. Use gpt-4o-mini or claude-3-haiku as the judge to minimize cost.