education-ai-guardrails

Use this skill when designing, reviewing, or drafting AI features for education, tutoring, grading, student support, learning analytics, classroom tools, curriculum content, academic integrity, child

SKILL.md

GitHub

Education AI Guardrails

Build education AI as teacher/student support with privacy, accessibility, age-appropriate design, and human oversight.

Workflow

  1. Classify the context: tutoring, lesson planning, grading feedback, assessment generation, accessibility support, student analytics, parent communication, or admin workflow.
  2. Identify learners: age range, school/university/workplace setting, accessibility needs, language needs, and whether minors are involved.
  3. Define the AI role: explain, draft, practice, summarize, adapt content, flag risk, or support an educator. Avoid unreviewed final grading or disciplinary decisions.
  4. Check data boundaries: student records, personal data, consent, retention, vendor sharing, classroom recordings, and local policy requirements.
  5. Validate pedagogy: learning objective alignment, source accuracy, age appropriateness, bias, hallucination risk, and accommodations.
  6. Add human oversight: teacher review for grading, curriculum, interventions, escalations, and any high-impact student outcome.
  7. Add misuse controls: plagiarism expectations, prompt-injection resistance, cyber/safety content boundaries, and audit logs for sensitive workflows.
  8. Deliver a risk-based recommendation with sources checked, human owner, validation path, and unresolved policy questions.

Checklist

  • Use official curriculum, school policy, or educator-provided materials as the source of truth when available.
  • Make feedback formative and explainable; avoid opaque scores without evidence.
  • Provide accessibility alternatives and support for multilingual learners.
  • Avoid collecting more student data than the feature needs.
  • Include safe escalation for self-harm, abuse, violence, or safeguarding signals.
  • Test outputs across diverse student profiles and edge cases before classroom rollout.

Guardrails

  • Do not make final grades, placement, discipline, special-education, or safeguarding decisions without authorized human review.
  • Do not expose student records, minors' data, or classroom recordings to unapproved tools.
  • Do not present AI-generated educational content as policy- or teacher-approved unless it was reviewed.
  • Do not help students cheat; redirect to learning support, citation, and integrity-preserving guidance.