Learn with Tutor Skills¶

The handbook is large — 16 core courses, 3 optional tracks, deep dives, projects, and interview prep. You don't need to read linearly.

This repo includes Claude Code / Cursor Agent Skills that act as tutors: tell the agent your goal, and it routes you to the right pages, plans your session, and teaches from handbook content.

Quick start¶

Clone this repo and open it in Claude Code or Cursor.
Skills load automatically from .claude/skills/ (see Skills & Rules).
State your goal in chat — no special syntax required.

Try these prompts

I'm a backend engineer — I want to ship a RAG app in a month. Where do I start?

Explain self-attention like I'm prepping for an interview.

I have 90 minutes today. Coach my study session on agents.

Quiz me on RAG chunking — I just read course 06 lesson 3.

Three learner skills¶

Skill	You say…	You get…
learning-path-advisor	"I want to learn …" / "Where should I start?" / "Help me get a job"	Multi-week path with exact `docs/` links, prerequisites, and a build project
ai-tutor	"Explain …" / "I don't understand …" / "Quiz me on …"	Socratic explanation grounded in handbook lessons + links to go deeper
study-session-coach	"I have 2 hours" / "Coach my session" / "Plan my week"	Time-boxed plan: which file to open, what artifact to produce, end-of-session check

Skills live in:

.claude/skills/
  learning-path-advisor/SKILL.md
  ai-tutor/SKILL.md
  study-session-coach/SKILL.md
  references/handbook-routing.md   ← shared routing index

How routing works¶

flowchart LR
  U[You state a goal] --> R{Skill router}
  R -->|path / career| A[learning-path-advisor]
  R -->|explain / quiz| T[ai-tutor]
  R -->|today / this week| C[study-session-coach]
  A --> H[Handbook docs/]
  T --> H
  C --> H
  H --> P[Project or eval artifact]

Each skill reads existing handbook files before responding:

docs/start-here.md — personas and goals
docs/learn/study-plans.md — week-by-week schedules
docs/topic-map.md — concept lookup
Course index.md and lessons/*.md — exact next steps

The agent does not invent a parallel curriculum — it navigates what you already have.

Pick the right skill¶

Situation	Use
Don't know where to begin	learning-path-advisor
Know the topic area, need schedule	learning-path-advisor → then study-session-coach
Reading a lesson, concept is fuzzy	ai-tutor (paste the lesson path if you can)
Finished a lesson, want retention	ai-tutor in quiz mode
Interview in 2 weeks	learning-path-advisor (interview goal) + study-session-coach (daily drills)

Example session flow¶

Week 0 — plan

"I'm an ML engineer, new to agents, 6 hrs/week for 8 weeks."

→ learning-path-advisor returns the agent-engineer study plan with links to Agent Engineering + courses 07–08.

Week 1 — day 1

"Coach a 90-minute session starting at the agent loop."

→ study-session-coach assigns The Agent Loop + artifact: implement the minimal loop from the lesson.

Week 1 — day 2 (stuck)

"Why does my agent loop never terminate?"

→ ai-tutor explains termination + links to Harness Engineering.

Study habits (built into the coach)¶

From Study Plans:

One lesson → one artifact — script, notebook, or eval case
Teach it — explain the concept aloud in 5 minutes after reading
Use the glossary — Glossary when terms blur

Portfolio rule: ship one RAG app, one agent, one production demo with evals — see Build These First.

Cursor vs Claude Code¶

Both load skills from .claude/skills/ when this repo is open.

Tool	Notes
Claude Code	Reads root `CLAUDE.md` every session — points to tutor skills
Cursor	Agent Skills auto-discover from skill `description` field — see Cursor Skills docs

Skill descriptions include trigger phrases ("Use when user wants to learn…") so the agent picks the right tutor without you naming the skill file.

Contributing tutor improvements¶

To improve routing or add trigger phrases, edit the skill's SKILL.md or .claude/references/handbook-routing.md. Authoring conventions: Skills & Rules · Contribute

Next: Start Here · Study Plans · Topic Map