What problem does it solve? Student names must never reach a cloud LLM, but instructors still need AI help processing submissions and grades. This Skill strips identity from student artifacts into opaque keys before AI processing and maps keyed results back to real students locally afterward. ## Core Features & Use Cases - Course-wide de-id master: Builds a .deid_master.csv with one row per student keyed by user_id, using a stable opaque deid_code = sha256(user_id)[:6], deduplicating multi-section enrollments. - Format-specific de-identification: Adapters handle Word docs, PDFs, Excel files, Jupyter notebooks, Databricks exports, plain text, Canvas submission comments, and gradebook exports. - Key-based re-identification: Maps keys back to user_id via .keymap.json with duplicate-aware handling, never by row position or sort order. - Leak verification: Scans de-identified artifacts against .known_names.txt to catch any name that slipped through the scrub. - Use Case: Before sending a batch of student essays to an LLM for feedback drafting, de-identify the submissions, run the AI analysis on opaque keys, then re-identify the keyed feedback into a named report stored outside the repo. ## Quick Start Ask the agent to rebuild the de-id master and de-identify the submissions in the current grading batch before any AI processing.
