How Educators Can Use Claude Skills to Turn a Rubric and Student Drafts Into Consistent Feedback Comments

If Claude Skills is new to you

Claude Skills are reusable folders that Claude can load when a task matches the Skill description. A custom Skill usually includes a Skill.md file with metadata, instructions, examples, and optional supporting files. Anthropic's current help says Skills are available on Free, Pro, Max, Team, and Enterprise plans when code execution is enabled.

For this workflow, you do not need to become a general Skills expert first. You will create one private Skill, upload it in Claude under Customize > Skills, enable it, test whether Claude recognizes it, and then use it on one draft at a time. The Claude Skills tool hub is useful further reading after you understand this page's workflow.

Problem and who this is for

Giving writing feedback takes time, and the real difficulty is consistency. The first paper gets detailed comments. The eighth paper often gets shorter comments because your attention is lower. A reusable Claude Skill can help by keeping the feedback structure stable while you stay responsible for the grade, final comments, and any sensitive wording.

This workflow is for teachers, faculty, tutors, and writing coaches who already use a rubric and want criterion-level feedback comments that are more even across a stack of drafts. It is not a full assessment automation system. The goal is a reviewed comment draft, not automatic grading.

What skill you will use in this workflow

You will build a private custom Skill called rubric-feedback-comments. The Skill tells Claude how to read the rubric first, connect comments to specific criteria, extract evidence from the student draft, and avoid assigning grades automatically.

This task is repetitive enough to deserve a Skill because the same educator often wants the same comment pattern across many drafts: strength, growth area, next revision step, and a clear reminder when evidence is weak or missing. Before using the Skill, prepare the assignment prompt, rubric, one student draft, and a short note about your preferred feedback tone.

Prerequisites

• Claude with Skills available and code execution enabled.
• Permission from your school, department, or client to use Claude with the student work in question.
• The assignment brief and rubric in text, DOCX, or PDF.
• Student drafts in DOCX, PDF, or pasted text. If you only have scans, run OCR first so the text is selectable.
• A short note on your preferred feedback voice. Example: concise, direct, respectful, and focused on the next revision.
• A plan to review every comment before sharing it with a student.

How to gather the source material

1. Save the rubric as a clean text file or DOCX. If it only exists in a slideshow or scan, copy it into a simple document first.
2. Save the assignment prompt in the same folder so Claude can distinguish rubric criteria from task requirements.
3. Convert any locked or image-only PDFs into searchable text before you start. If you can highlight the words, Claude will have a much easier time.
4. Create one small file called feedback-style.txt with the comment style you want. Example: one strength, one growth comment, one next revision step, and no grade language.
5. Choose one sample draft for the first trigger test. Do not start with the whole stack.
6. Work one student draft at a time unless your institution has explicitly approved a different batch process.

Build the custom skill for this workflow

Create a folder named rubric-feedback-comments. Inside it, create a file named Skill.md. Paste the starter content below into that file, then adjust the wording for your course, school, age group, and feedback style.

After the folder is ready, zip the rubric-feedback-comments folder itself so the ZIP contains the skill folder at its root. Do not zip loose files by themselves. In Claude, go to Customize > Skills, upload the ZIP file, and enable the Skill. Then run the trigger test below before using it on real student work.

---
name: rubric-feedback-comments
description: Compare student work against a rubric, extract criterion-level evidence, and draft consistent feedback comments without automatically assigning grades.
---

# Rubric Feedback Comments

Use this skill when the user wants criterion-level feedback comments from a rubric and one or more student drafts.

## What this skill should do
- Read the rubric first.
- Match each comment to a specific criterion.
- Cite short evidence from the student work where helpful.
- Draft clear, supportive comments the teacher can review and revise.

## Output shape
- One section per rubric criterion.
- Strength comment
- Growth comment
- Suggested next revision step

## Guardrails
- Do not assign grades automatically unless the user explicitly asks and the teacher has approved that use.
- Do not overstate evidence.
- Keep comments respectful and specific.
- If evidence is missing, say so plainly.

## Examples
- Use this rubric and this student draft to draft criterion-level feedback comments.
- Generate review comments that match the rubric without assigning a grade.
- Apply my rubric feedback workflow to this essay.

Treat this as a starter template, not the perfect final Skill. If your rubric always has specific criterion names, add them. If your school requires certain language around accommodations, late work, or academic integrity, add those boundaries too.

Skill-creator helper

Anthropic's official skill-creator Skill can help draft or refine a private Skill like this. Use it as a helper when you want better trigger wording, clearer guardrails, or sample test prompts. Keep the final Skill private if it contains school-specific rubric language or feedback policies.

Why this should be a custom skill, not a one-off prompt

A one-off prompt works once, but feedback work repeats. A Skill keeps the same evidence rules, comment sections, tone expectations, and no-auto-grade guardrail available every time you review a draft. That reduces prompt rewriting and makes it easier to keep comments consistent across students while still leaving the final judgment with the educator.

Skill trigger test

After enabling the Skill, test it with one sample draft before using it on a larger set. These prompts should cause Claude to use the rubric-feedback-comments Skill:

• Use this rubric and sample draft to create criterion-level feedback comments.
• Apply my feedback workflow without assigning a grade.
• Draft comments tied to the rubric and keep weak evidence visible.
• Use the rubric feedback Skill on this essay and show missing evidence plainly.

If Claude does not appear to use the Skill, tighten the Skill description and test again with language that names rubric feedback, criterion-level comments, and no automatic grade.

Numbered workflow steps

1. Gather and clean the source material

Put the assignment prompt, rubric, feedback-style.txt, and one student draft in a clean workspace. Remove student names if that is required by your policy. Make sure the rubric criteria are readable and that the student draft text is selectable.

Do not ask Claude to infer the rubric from the assignment prompt. Give the rubric directly, even if it is short.

2. Create or enable the custom Skill for this workflow

If this is your first run, build the rubric-feedback-comments folder, add the Skill.md file, zip the folder, upload it in Customize > Skills, and enable it. If you already built the Skill, open Claude's Skills settings and confirm it is enabled before you start.

Keep the Skill focused on criterion-level feedback comments. Do not add grading formulas, grade conversion tables, or classwide analytics unless your institution has approved that use and you actually need it.

3. Run a quick trigger test with one sample draft

Attach or paste the rubric and one sample draft. Use a prompt like:

Use this rubric and sample draft to create criterion-level feedback comments. Do not assign a grade. Keep weak or missing evidence visible.

Review whether Claude follows the Skill: one section per rubric criterion, a strength comment, a growth comment, a next revision step, and no automatic grade. If the output is generic, add more concrete tone guidance to feedback-style.txt or tighten the Skill guardrails.

4. Run the actual feedback workflow with the Skill active

For each student draft, attach the assignment prompt, rubric, feedback-style.txt, and the draft. Ask Claude to use the rubric feedback workflow and keep comments evidence-based.

Use one draft at a time for normal feedback work. This makes it easier to catch places where Claude overstates evidence, misses nuance, or writes comments that do not sound like you.

5. Do a human review for accuracy, boundaries, and tone

Before you share anything, review every criterion section. Check that each comment matches the rubric, the cited evidence is actually present, and the suggested next step is realistic for the student.

You remain responsible for grading decisions and final comments. The Skill should not assign grades automatically, invent evidence, or turn an ambiguous passage into a confident judgment.

6. Save, send, or reuse the result

Paste the reviewed comments into your LMS, feedback document, or gradebook workflow. Save the final version near the source draft if your school policy allows it. After several runs, update the Skill with wording that consistently improves the output, such as preferred transition phrases or criterion labels.

Tool-specific instructions

Use Claude Skills for the reusable parts of the task: rubric-reading order, evidence extraction rules, comment structure, tone expectations, and the no-auto-grade guardrail. The page-specific Skill should come first because this workflow depends on your rubric, your feedback style, and your institution's rules.

For broader setup context, see the Claude Skills hub. For help improving the Skill itself, use Anthropic's official skill-creator as a drafting aid rather than a replacement for your private rubric Skill.

Quality checks

1. Each major comment points to evidence in the draft, not generic writing advice.
2. The output stays at the criterion-comment level and does not drift into full assessment automation.
3. The tone matches how you actually talk to students.
4. The comments do not quietly assign a grade or imply a final score.
5. Revision priorities are limited enough that a student could act on them.
6. Missing evidence, unclear claims, or uncertain rubric matches are visible instead of smoothed over.

Common failure modes and fixes

The comments sound generic

Fix: add a more specific feedback-style.txt file and require one concrete evidence reference under each criterion before the final comments.

Claude overstates a weakness

Fix: add a guardrail to the Skill that every growth comment must be supported by a short quote, paragraph reference, or clear description from the draft.

Claude starts grading automatically

Fix: tighten the Skill description and guardrails so the Skill is for feedback comments only. Add: "Do not assign grades, scores, levels, or bands unless the teacher explicitly asks after review."

The rubric is too broad

Fix: rewrite the rubric into plainer criterion labels before using the workflow. If a criterion is broad, ask Claude to mark the evidence as tentative and leave the final interpretation to you.

The Skill does not trigger

Fix: update the Skill description so it includes the exact phrases you use in prompts, such as "rubric feedback," "criterion-level comments," "student draft," and "without assigning grades." Then run the trigger test again.