Colin Rundel

You can scaffold a grading template from a key solution repo, generate a per-question grading rubric with point allocation, and apply that rubric headlessly to student repositories as first-pass machine grading. All outputs are validated against a bundled JSON Schema and imported into the project's grading database.

They target instructors and teaching assistants in statistics and data science courses who grade student code assignments at scale. They suit educators already managing GitHub-based assignments through markermd and ghclass who want structured, reviewable machine pre-grading before human review.

You need an initialized markermd project created with markermd::init_project() or ghclass::org_grade_assignment(), containing a key (solution) repository and student repositories. Rubric application additionally requires a stored grading rubric; rubric scaffolding requires an existing stored grading template.

The apply-rubric skill reads each student repo's per-question sections, matches rubric item descriptions verbatim, records reasoning as private notes, writes a marks YAML, validates it against the bundled JSON Schema, and imports it via marks_import() for human review in mark().

During rubric scaffolding, you specify the assignment's total points and choose a distribution strategy. The skill proposes equal splits, content-weighted splits, or assignment-stated splits, then drafts concrete key-derived rubric items per question and writes validated rubric YAML into the grading database.