Skip to main content
Kemahiran.AI
BM
Open menu

Rubric Design and Moderation Assistant

Build a usable rubric and test descriptor clarity against sample responses without automating final grades.

Start here

Two ways to use this skill

Use it immediately without installation, or upload the package to save it as a reusable Gemini Skill.

Quick use

Use directly in Gemini

  1. Copy the complete instructions in one click.
  2. Paste them into a new Gemini conversation.
  3. Replace the variables and attach the required materials.

Reusable setup

Add to Gemini Skills

  1. Download the ZIP package below—do not unzip it.
  2. In Gemini, open Customize → Skills → Create → Upload.
  3. Select the ZIP, review its details, then create the Skill.
Download for Gemini Skills
Estimated time: 30 minutes

Package includes: README.md, SKILL.md, LICENSES.md

Prepare before starting

  • Learning outcomes and task
  • Approved marking policy
  • De-identified sample responses when available

If the Skills menu is not yet available on your account, use Copy instructions. Gemini Skills availability may still vary by account type and staged rollout. Official Gemini Skills guide(opens in a new tab)

Complete guide

What you will get

A draft analytic rubric with observable criteria, meaningfully distinct performance levels and moderation notes based on sample responses.

Before you begin

Prepare the learning outcomes, task instructions, learner level, marking policy, approved rubric structure and at least two de-identified sample responses when moderation is required.

How to use it

  1. Supply the learning outcomes, task instructions, learner level and marking policy.
  2. Define criteria observable in the work.
  3. Use performance levels and marks only when approved.
  4. Draft descriptors that distinguish work quality rather than learner traits.
  5. Trial the rubric on at least two de-identified sample responses.
  6. Record overlaps, unassessable evidence and teacher decisions.

Copyable instruction

You help a teacher design and moderate a rubric. Use only the learning outcomes, task, policy and sample responses I supply. If the outcomes, task or rubric structure is missing, ask and stop.

Learning outcomes: {{LEARNING_OUTCOMES}}
Task: {{TASK}}
Learner level: {{LEVEL}}
Approved structure and marks: {{RUBRIC_STRUCTURE}}
Policy/guidance: {{POLICY}}
De-identified sample responses: {{SAMPLE_RESPONSES}}

Create a criteria-by-performance-level table using observable evidence. Avoid vague words such as “good” without explanation. Do not add outcomes, weights or penalties. If samples are available, apply the rubric tentatively, cite evidence and flag interpretation differences. Finish with fairness, accessibility, descriptor-overlap and teacher-decision checks. Do not issue final grades for real learners.

Worked example

Synthetic example: a four-level science-presentation rubric for factual accuracy, evidence and explanatory clarity. Two de-identified sample responses are used to identify overlapping or unobservable descriptors.

Quality check

  • Every criterion maps to a learning outcome.
  • Every level is meaningfully distinct and uses observable evidence.
  • Descriptors do not penalise language style unrelated to the outcome.
  • Marks, weights and penalties match the approved policy.

Limitations and responsible use

An AI-generated rubric is not validated for fairness or reliability. A teacher must verify alignment, language, accessibility and marking decisions.