What’s inside

One SKILL.md (the playbook) + one starter template. A preview:

Grading Assistant

When to use

• You have 30+ essays to grade against a rubric and the weekend is disappearing
• Every student should get real feedback, not a score and a "good job"
• You want to spot what the whole class got wrong and turn it into next week's review
• You teach the same course across sections and need grading to stay consistent
• You need lesson plans or review slides built from actual student mistakes, not your memory of them

Workflow

Set up one AI project per course and feed it the syllabus, your rubric, past exams, model answers, and course materials. One real case keeps a separate project per course so the grading stays anchored to that course's standards. A rubric written for a colleague, not for yourself, is the bar — the AI grades what it says, not what you meant.

1. Build the grading project (input: syllabus, rubric, sample answers → output: per-course project)

Strip names before grading: student papers get anonymous numbers. This keeps the grading honest and protects student privacy. Feed readable digital files — scans need to be legible or the grading degrades silently.

1. Anonymize the submissions (input: student work → output: numbered, ID-free batch)

The AI scores each submission against every rubric criterion separately, with a short justification per criterion. It doesn't invent the rubric as it goes — every score points at a criterion. Borderline and low-confidence grades get flagged for your review; the easy agreements flow through.

1. Grade criterion by criterion (input: rubric + submissions → output: scored batch + flagged borderlines)

Each student gets personalized comments: what they did well (citing their actual words), what cost them points (citing the criterion), and one concrete thing to fix next time. A feedback comment that could apply to any student applies to none of them — send it back.

1. Write real feedback (input: graded batch → output: per-student feedback)

Aggregate the criterion scores across the class to find the hot spots: the two rubric rows where everyone bled points. Turn those into next week's review — slides or a lesson plan built from the actual wrong answers, not from what you remember being hard. Never touch the original submission files; scores live in a separate sheet.

1. Mine the mistakes for teaching (input: full batch results → output: class trends + review materials)

Review the flagged borderlines, spot-check a handful, then approve. The AI proposes; you dispose. Educational responsibility doesn't delegate — final grades are always the teacher's call.

1. You set the final grades (input: scored batch → output: confirmed grades)

Output spec

-

• Per submission: score per rubric criterion with one-line justification each

Download the full skill pack (ZIP)