Video summary

Learn anything with the /teach skill

Main summary

Key takeaways

Technology

Overview

The video demonstrates the creator’s “teach” skill (from their skills repo) and argues for a stateful, structured tutoring workflow powered by AI.

Core technological concept: stateful vs. stateless skills

Stateless skills

  • No memory across runs
  • Don’t save progress to disk or to MCP servers
  • Example: “grill me” (repeatedly quizzes until the learner is ready)

Stateful skills

  • Save progress and learning artifacts locally (and/or to MCP servers)
  • Improve over time
  • Example: “grill with docs”, which saves items like ADRs (architectural decision records), glossaries, etc.

Why “teach” matters

  • The teach skill is explicitly designed to be stateful
  • It remembers where the learner is and what to teach next

How to install and use the teach skill

Installation

  • From the creator’s skills repo at mapper.got skills
  • Use the skills.sh installer
  • Select the Teach skill

Usage

  • Run Teach in an empty directory inside your coding agent/workspace
    • Example: a Rubik’s Cube project folder
  • The skill generates a full set of learning files and updates them as you progress

What the teach skill produces (workspace artifacts)

  1. Mission file (mission.md)

    • Defines a learner goal and the reason for learning
    • Example goal: solve a scrambled 3x3 Rubik’s Cube unaided at least once, focusing on the achievement (not speed/theory)

  2. Web-sourced high-trust resources

    • On the first run, it searches for primary, high-quality resources
    • Continues updating these resources as learning proceeds

  3. Lessons folder

    • Each lesson is saved as an individual HTML file
    • HTML is highlighted as enabling richer, more interactive lessons than Markdown
    • First lesson example: “anatomy, notation and the white cross”
    • Lesson components may include:
      • Diagrams
      • Simple explainers
      • Callouts
      • Quizzes/feedback

  4. Learning records

    • After you confirm progress (e.g., “Yes, I can make the white cross”), the agent records it
    • On subsequent Teach runs, it uses these records to:
      • Diagnose gaps
      • Tailor the next lesson

  5. Community / research prompts (wisdom delegation)

    • Tries to identify communities where the learner can ask questions
    • Rationale: materials provide knowledge/skills, but “wisdom” comes from community interaction and real-world practice

  6. Reference material

    • Glossary for terminology (e.g., notation, jargon)
      • Helps future lessons be more concise via glossary references
    • Cheat sheets (e.g., a “solve card” with the full solve)
    • notes.md for internal teacher notes (preferences, “watch outs,” etc.)

Pedagogy: adaptive tutoring in the learner’s “zone of proximal development”

The skill uses the teaching idea zone of proximal development:

  • Teach at a point where the learner is challenged but not intimidated
  • Produce lessons that are concise, compact, and tightly framed

Example second-session behavior

  • It checks existing saved artifacts (solve cards, progress)
  • Diagnoses the learner understands the corner cycle concept but lacks muscle memory
  • Generates a targeted next lesson accordingly

Interactive lesson experience (HTML advantages)

  • Lessons can include interactive/tappable guidance
  • There’s a guided mode you can toggle
  • The speaker emphasizes HTML’s access to browser capabilities to make practice interactive

Engineering / skill design summary (what the skill does)

  • The user request is stateful: learning is intended to span multiple sessions
  • Internally, it has a structured “philosophy”:
    • Knowledge from high-trust resources
    • Skills from interactive lessons and practice
    • Wisdom via delegation to communities
  • The creator frames teach as a way to avoid dependence on the agent permanently—helping the learner become confident enough to engage with communities independently

Stated use-case beyond Rubik’s Cubes

The creator suggests it could be valuable for onboarding to codebases:

  • Documentation is often outside the learner’s immediate zone of proximal development
  • Teach could start learners in their own workspace, point them at a codebase, and teach:
    • concept/problem-domain first
  • Result: faster movement toward productivity

Tech / model mentioned

  • Opus 4.8
  • Run with medium effort

Main speakers / sources

  • Speaker: the creator of the skills/teach system (references “my skills repo” and “my Teach skill”)
  • Sources referenced:
    • The skills repo at mapper.got skills
    • Examples from prior skills: “grill me” and “grill with docs.”

Original video