Content
Cumulative packages across four levels; in person, one level corresponds to one training day, while self-study participants cover the same level asynchronously.
Level 1: Augmented Foundations
7 lessons
The user layer: commands, memory, MCP servers, reusable skills, prompting and git workflows.
Included in: Master of AI-Augmented Coding, Master of AI-Augmented Orchestration, Master of AI-Augmented Architecture
- 1.1
Foundational Concepts & The Current Landscape
What is Claude Code, the agentic paradigm, tokens & context, current models, effort and cost, the mindset shift from coder to orchestrator.
- Challenge: Estimate Your Token Budget
- 1.2
Repository Introduction: Meet invoices-glider
Repository Introduction: Meet invoices-glider · What invoices-glider Is · Mini Exercise: Get Oriented in invoices-glider
- Challenge: Get Oriented in invoices-glider
- 1.3
Prompting Practice
The CLEAR prompt (Context, Limits, Evidence, Aim, Result), based on Anthropic's prompting guide, practiced on invoices-glider: vague vs CLEAR runs, three starter prompts, and a two-way anonymizer for working with sensitive data.
- Challenge: Vague vs CLEAR
- Challenge: Starter Prompt: Build
- Challenge: Use Claude Without Showing It the Real Data
- 1.4
Commands & Memory
Steering a running session, operating modes, slash commands, CLAUDE.md and file imports.
- Challenge: Create Your CLAUDE.md File
- Challenge: Prove It: Is Your Rule Read?
- 1.5
MCP Servers: Extending Claude's Reach
Model Context Protocol: what it is, plugins and marketplaces, installing servers, and why to connect only what a task needs.
- Challenge: Connect Your First MCP Server
- 1.6
Skills Hands-on, Part 1: Write Your Own and Run Ours
Auto-discovered workflows, SKILL.md structure, triggers, when skills make sense vs commands.
- Challenge: Write Your Own Skill
- 1.7
Skills Hands-on, Part 2: Composition, Subagents and Limits
Skills: Composition, Subagents & Limits · Claude Is Already Orchestrating · A Skill That Dispatches a Subagent
- Challenge: Build a Dispatcher
- Challenge: Cap the Depth
Level 2: Augmented Coding
6 lessons
The system layer and the first orchestrator.
Included in: Master of AI-Augmented Coding, Master of AI-Augmented Orchestration, Master of AI-Augmented Architecture
- 2.1
Skills: Advanced (Hands-on Opener)
Auto-discovered workflows, SKILL.md structure, triggers, when skills make sense vs commands.
- Challenge: Create a Skill
- 2.2
Subagents
Built-in and custom subagents, background and fork mode, the orchestrator-worker pattern, scoped tools per worker.
- 2.3
The Skill / Agent / Hook Architecture
Commands vs. Skills vs. Subagents vs. Hooks. When to use each. The orchestrator pattern.
- 2.4
Build Your First Subagent + Safety Hook
Practical exercise: Create a personalized unified development workflow with enforcement thinking.
- Challenge: Ship a Subagent
- Challenge: Ship One Safety Hook
- 2.5
Your First Orchestrator, Ship Through It, Then Into Your Own Repo
Compose your first orchestrator on the practice repo (a researcher leaf finds the bug, backend-engineer fixes it, reviewer and tests check it), ship one feature through it, then take protected time to port your setup into your own repo with /port-to-your-repo and prove it fires there.
- 2.6
The /insights Command
Built-in command that analyzes your Claude Code sessions and generates personalized recommendations.
Level 3: Augmented Orchestration
7 lessons
Developing multi-level orchestrators with self-repair loops.
Included in: Master of AI-Augmented Orchestration, Master of AI-Augmented Architecture
- 3.1
Warm-Up: Run What You Already Have
Warm-Up · You Start With a Working Orchestrator · Mini Exercise: Ship a Small Feature
- Challenge: Ship a Small Feature
- 3.2
The Three-Layer Model + The Gate-Check
Three Layers, One Folder · What Each Layer Does · The Isolation Levels, and What Changed Since June
- Challenge: Gate-Check Three Real Tasks
- Challenge: Name Your Three Layers
- 3.3
Build a More Complex Orchestrator
Three Clean Layers: a Choice, Not a Constraint · The Real .claude/ Tree You Are Building · The L2 Dispatcher: Decide, Then Fan Out
- Challenge: Grow /implement Into a Three-Layer Orchestrator
- Challenge: Run It, Break It, Commit It
- 3.4
The Boundary Is Now a Choice (Nested Subagents)
The Boundary Is Now a Choice · The Rule You Learned Was a Constraint · One Axis: Inline vs Firewall
- Challenge: Build a Firewall, Then Cap the Depth
- 3.5
Parallel Execution: Git Worktrees & isolation: worktree
Why Parallel Work Needs Isolation · The Raw Mechanics: git worktree · The Frontmatter: isolation: worktree
- Challenge: Fan a Writer Leaf Across 5 Files
- Challenge: Dispatch the Fan-Out, Then Merge Back
- 3.6
Self-Correction Loops + Structural Requirements
Where the Loop Lives: L1 Owns It · Three Ways to Verify · Contracts That Make It Converge
- Challenge: Build the L1 Repair Loop
- Challenge: Build the Loop and Run It
- 3.7
Apply It: Run the Whole Machine
Apply It · Mini Exercise: Run the Whole Machine · Mini Exercise: Watch the Chain
- Challenge: Run the Whole Machine
- Challenge: Watch the Chain
Level 4: Augmented Company
11 lessons
Advanced techniques and team infrastructure: patterns beyond the orchestrator (convergent loop, pipeline, generator–critic, swarm, workflows), distribution of a custom setup through plugins and a private marketplace, governance and enforcement in an AI-augmented team, and abstraction over data and code through a bounded team MCP server.
Included in: Master of AI-Augmented Architecture
- 4.1
Let It Run Safely (Read-Only DB)
Let It Run Safely · Hands-on: The Database Is Read-Only Now · Why You Can Let It Run Unattended
- Challenge: Try It
- Challenge: Then Make It Yours
- 4.2
Micro-Tools
Small HTML pages Claude builds so people can review items and decide; Claude reads the decisions back and acts.
- Challenge: Triage Docs vs Code
- 4.3
Build Your Own MCP
Build an MCP server that talks to the running invoices app through its API, then connect your agent to it.
- Challenge: Add a Tool: What Does a Client Owe?
- Challenge: Install It in Claude Code
- Challenge: Use It
- 4.4
RAG & Vector Stores
Agentic search vs retrieval, how retrieval works, a hands-on with station's docs-search, and what RAG costs.
- Challenge: Plug In docs-search
- Challenge: Ask What grep Cannot Find
- 4.5
Workflows and Loops, Part 1: Workflows
Workflows (New Platform Feature) · A Workflow Is a Script Claude Writes · Run It, Watch It, Save It
- 4.6
Workflows and Loops, Part 2: The Convergent Loop
Advanced Use Cases: The Convergent Loop · Loops That Know When to Stop: /goal · Same Loop, Different Done-Condition
- Challenge: Run a Convergent Loop
- 4.7
Design-agent-system
What /design-agent-system Does · The Plan It Returns · Mini Exercise: Run It On Your Repo
- Challenge: Run It On Your Repo
- Challenge: Review the Map
- 4.8
Distributing Your Setup
What a Plugin Bundles · What Travels, What Does Not · A Private Marketplace Is a Folder or a Git Repo
- 4.9
Architecture, Platform Features & Folder Layout
Memory Tiers, Mapped to Files · 25 Skills: Flat Discovery vs. a Readable Tree · Regenerate the Flat Symlinks
- Challenge: Flat → Source + Symlink
- Challenge: Regenerate and Confirm
- 4.10
Knowledge Base & Artifacts
Turn what the team knows into memory the agent loads: the memory tiers and their files, what stays out, a docs/ knowledge base of decisions and incidents that loads when the work touches it, and how to keep it true. Hands-on: give station a team memory and prove the agent uses it.
- Challenge: Open Station's Docs in Obsidian
- 4.11
Capstone: After Today
Write a 30-day rollout plan for your real team from today's pieces: what to promote first and through which gate, which memory to seed, which guardrails to set, which signals to watch, who owns what and your first champions. Then swap plans and find the riskiest assumption.