What problem does it solve? Grading handwritten university exams for large classes is slow, error-prone, and hard to audit. This Skill standardizes the entire pipeline—scanning, identity matching, parallel AI grading, machine verification, teacher review, return, and appeal handling—so a 130-student exam can be graded and returned the same day with full traceability. ## Core Features & Use Cases - End-to-end grading workflow: Covers exam design, scan pairing, student ID OCR, rubric definition, parallel AI grading agents, and mechanical score verification with Python scripts. - Mandatory teacher review HTML: Generates a single HTML file embedding answer image crops so instructors can adjudicate ambiguous readings and out-of-rubric partial credit before scores are finalized. - Prompt injection defense: Built-in countermeasures against students writing instructions to the grading AI on their answer sheets, with detection logging. - Return and appeals support: Produces per-student annotated PDFs, Excel gradebooks, and Google Drive/Classroom return scripts with documented accident-prevention rules. - Use Case: A professor scans 132 handwritten final exams, runs 22 parallel grading agents against a documented rubric, verifies totals programmatically, reviews 13 flagged cases via image-embedded HTML, and returns personalized graded PDFs the same day. ## Quick Start Tell Claude you want to grade a written exam and it will walk you through rubric creation, scanning, parallel AI grading, teacher review, and returning results to students.
