GarethManning/education-agent-skills

What it solves

Most AI education tools are built on convention rather than evidence, risking the scale-up of ineffective teaching. This project provides a credible, research-anchored foundation for AI in education — a set of reusable skills that help agents produce lesson plans, assessments, curriculum designs, and study support grounded in named research and clearly labelled practitioner frameworks. It also addresses the workload problem: educators can get structured, evidence-based support in minutes instead of hours of research.

How it works

The library uses the open Agent Skills format (originally developed by Anthropic). Each skill is a folder containing a SKILL.md with instructions, plus optional supporting resources. Skills are grouped into 20 domains: 19 for teachers and designers (e.g., memory & learning science, curriculum design, inclusive design) and 1 for student-facing study interactions. Every skill includes YAML schema headers, typed inputs and outputs, and chaining metadata, making them programmatically usable. The library ships as canonical source files, a portable edition that passes the official format validator, and integrations for Claude, Codex, and Hermes. You can also copy a skill's prompt manually and paste it into any AI.

Who it’s for

• Classroom teachers wanting evidence-based lesson and assessment design without hours of research
• University lecturers who lack formal teacher training and want practical, research-grounded support
• Curriculum designers and heads of learning building programmes and units
• School leaders in innovative and alternative education contexts (international, Montessori, project-based, etc.)
• Innovators reimagining education and building new school models
• EdTech developers and AI builders who need a structured, programmatically accessible education knowledge layer
• Education researchers interested in how evidence translates into AI-mediated practice

Highlights

• 165 skills across 20 domains, covering everything from retrieval practice to student-facing study coaching
• Evidence strength is rated transparently: Strong, Moderate, Emerging, or Original (practitioner frameworks clearly labelled)
• Excludes poorly evidenced approaches like learning styles and VAK, with an explicit exclusion list
• Built by an educator with 20 years of international school experience, with limitations and educator judgement built into every prompt
• Designed for orchestration from day one — typed inputs/outputs, chaining metadata, and composable outputs
• Multiple integration routes: Claude plugin, Codex CLI, Hermes tap, portable ZIP, or manual copy-paste