- Copy
.claude/skills/rubric-feedback/references/rubrics/audience.mdas a template. - Fill in each section:
- Rubric definition — exact wording the school uses (cite the source).
- Strong example — body application — a concrete passage that demonstrates the rubric.
- Strong example — contextualization footnote — the footnote for that passage.
- Why this pair works — explain why the example is strong.
- Evaluation criteria — bullet-level criteria for body and footnote separately.
- Common weaknesses — table with weakness / what-it-looks-like / stronger-version.
- Add a row to
.claude/skills/rubric-feedback/references/rubrics/INDEX.mdunder the correct discipline section. - Open a PR. Every rubric definition must include a citation to its official Minerva source.
- Open
.claude/settings.json. - Copy one of the existing discipline agent blocks under
hooks.PreToolUse[0].hooks. - Update
prompt,statusMessage, and discipline-specific evaluation criteria. - Test with a sample PDF containing rubric tags from that discipline.
- Add partial-failure handling note if the subagent has unique failure modes.
- Rubric definition must cite its official source (Minerva course doc or LO registry).
- Strong examples must be original or anonymized — no real student work without consent.
- Cross-platform changes (Cursor rules, AGENTS.md) must mirror the SKILL.md workflow step-for-step.
- No placeholders (
TODO,TBD) in merged rubric files — placeholder files (NS/SS) are exempt until content is available.