HARNESSING / CLAUDE CODE

From zero to graph engineering on AI agents — for people who use them.

A free, open-source course. 119 lessons across 19 phases, 19 graded practice modules. MCP, hooks, plugins, worktree isolation, dynamic workflows, harness engineering, task graphs.

Start with Lesson 1 → See the whole map

Every abstraction in Claude Code resolves to something you can reason about, break, and rebuild. This course teaches the mechanism — not the prompt.

Why this course exists

Most agentic tooling is taught in two broken ways: as a list of features, or as prompts to copy. Neither teaches the mechanism. The same fear — losing work to an agent run — drives both failures. The fix is to know what the agent is doing. This course is that.

The 19 phases

Phase 00

Setup & Tooling

Your tools shape your thinking. Set them up once.

5 lessons · module 00.1 Break
Phase 01

Foundations of AI

The words everything else leans on.

6 lessons · module 01.1 Instrument
Phase 02

Git Deep

Commit = photograph. Branch = label. `worktree` hides `working tree`.

7 lessons · module 02.1 Break
Phase 03

Models & Families

Open-weight vs frontier. What actually costs money.

8 lessons · module 03.1 Instrument
Phase 04

Claude Code

The tool, from first run to fluency.

11 lessons · module 04.1 Build
Phase 05

Commands & Skills

The most flexible extension layer.

7 lessons · module 05.1 Break
Phase 06

Subagents

Context isolation.

4 lessons · module 06.1 Build
Phase 07

MCP

A protocol, not a tool.

6 lessons · module 07.1 Decide
Phase 08

Hooks

Events on the loop. Enforcement, not requests.

6 lessons · module 08.1 Build
Phase 09

Plugins

The packaging layer. Comes last for a reason.

4 lessons · module 09.1 Build
Phase 10

Memory & Instructions

How context persists.

6 lessons · module 10.1 Build
Phase 11

Parallelism I

The five names, and why they're different.

6 lessons · module 11.1 Build
Phase 12

Parallelism II

Orchestration.

9 lessons · module 12.1 Break
Phase 13

Context Engineering

Context as architecture.

5 lessons · module 13.1 Instrument
Phase 14

Harness Engineering

The loop minus the intelligence.

7 lessons · module 14.1 Instrument
Phase 15

Graph Engineering

Nodes, edges, and what genuinely parallelizable means.

8 lessons · module 15.1 Decide
Phase 16

Worktrees in Practice

Deep on the one you came for.

5 lessons · module 16.1 Break
Phase 17

Scaling & Operations

Running it for real.

5 lessons · module 17.1 Build
Phase 18

Capstone

Ship it.

4 lessons · module 18.1 Build

Teaching it to yourself, with an agent

This repo is designed to be pointed at an AI agent. The MCP server and REST API are the agent entry points. Five skills, each invoked deliberately by the learner:

CommandDoes
/teach-meTeach one lesson, at the learner's actual level
/run-moduleStart or grade a practice module against its rubric
/where-am-iWhat's demonstrated, what's next
/write-lessonAuthor a new lesson to template
/run-workflowAuthor the corpus, validate it, rebuild this site