● SYSTEM BLUEPRINT / 32 LESSONS
Claude Code Architecture Mastery
A 32-session deep dive into how Claude Code actually works. From the agent loop to the Fable 5 era of models, memory, and workflow orchestration.
Your last session:
- Lessons
- 32 Sessions
- System layers
- 5 Modules
- Estimated time
- 8h 51m
CHOOSE THE RIGHT MAP
One ecosystem, five different questions
Use each section for a different decision. This course explains the internals; the maps show evolution and product coverage; the guides turn capability into work.
- 01 ↗ Learn
Find the right starting point and build a mental model.
Choose a path - 02 ↗ Evolution map
See what the agent toolchain has completed and what a full computer still needs.
Open the map - 03 ↗ Tool landscape
Compare Claude Code, Codex, and Grok by capability and direction.
Compare tools - 04 You are here Architecture course
Understand how Claude Code works beneath the interface.
Study the system - 05 ↗ Guides
Apply the tools to concrete writing, analysis, coding, and business work.
Ship a task
CURRICULUM CONTROL DECK
Explore the system layer by layer
Search all 32 lessons or narrow the curriculum by system layer and difficulty.
32 lessons shown
M01 Module 1: Core Agent Reasoning · tools · context 6 sessions 1h 36m 2/6 Verified
Trace one complete agent loop and predict when tools, plans, and context change the result.
Reasoning · tools · context
- S01 The Agent Loop Understand the core async message loop that powers Claude Code — how it processes messages, routes tool calls, and decides when to stop.
- S02 Tool System & Permissions Learn how Claude Code turns model requests into tool actions, and how permission modes and rules gate those actions.
- S03 Planning with Task Tools Use Plan mode and Claude Code's structured Task tools to explore, approve, execute, and verify complex work.
- S04 Subagents & Context Isolation Learn how Claude Code spawns child agents with clean context windows — enabling parallel work, protecting the parent's context, and handling complex subtasks.
- S05 Skills & Knowledge Loading Understand Claude Code's two-layer knowledge injection — how skill names are loaded upfront while full definitions are deferred until needed.
- S06 Context Compaction Learn how Claude Code manages context window limits with a 3-layer compression strategy — keeping conversations productive even on long sessions.
M02 Module 2: Multi-Agent Orchestration · parallelism 6 sessions 1h 33m 2/6 Verified
Design parallel work that has explicit ownership, dependencies, isolation, and a merge path.
Orchestration · parallelism
- S07 Task Graph & Dependencies Model complex work as a dependency graph with TaskCreate, TaskUpdate, TaskGet, and TaskList.
- S08 Background Tasks Understand daemon threads and notification queues for long-running background operations.
- S09 Agent Teams & Communication Coordinate a lead and independent teammates through a shared task list and mailbox.
- S10 Team Protocols Master the request-response FSM with request_id for reliable multi-agent coordination.
- S11 Autonomous Agents Build agents with WORK/IDLE phases and auto-claim capabilities for independent operation.
- S12 Worktree Isolation Run parallel Claude Code sessions in separate Git worktrees and clean them up safely.
M03 Module 3: Real Architecture Protocol · memory · governance 6 sessions 1h 47m 4/6 Verified
Read Claude Code as a real system: protocol, extensions, persistence, policy, and security boundaries.
Protocol · memory · governance
- S13 Agent SDK Control Surface Build programmable agents with the public TypeScript Agent SDK and its query() stream.
- S14 MCP Integration Master the Model Context Protocol as a tool extension system.
- S15 Hooks System Automate and enforce Claude Code lifecycle rules with the current nested hooks schema.
- S16 Session Management Resume, name, branch, export, and automate Claude Code sessions through public interfaces.
- S17 CLAUDE.md Design Design concise, scoped project instructions that Claude Code loads into context.
- S18 Permission Models & Security Choose the right Claude Code permission mode, write precise rules, and combine permissions with sandboxing and policy controls.
M04 Module 4: Mastery Reliability · economics · oversight 6 sessions 1h 53m 0/6 Verified
Build an observable, recoverable workflow that can cross tools without losing control.
Reliability · economics · oversight
- S19 Multi-CLI Workflows Orchestrate parallel terminal sessions for maximum productivity.
- S20 Error Recovery Build retry logic, error classification, and fallback strategies.
- S21 Cost Optimization Optimize model selection, caching, and token management for efficiency.
- S22 Human-in-the-Loop Design confirmation flows, approval gates, and escalation patterns.
- S23 Custom Agents & Skills Build your own .md-based extensions for specialized workflows.
- S24 Production Patterns Master observability, monitoring, and deployment patterns for production use.
M05 Module 5: The Fable 5 Era Autonomy · continuity · evolution 8 sessions 2h 2m 4/8 Verified
Evaluate emerging capabilities without confusing product interfaces, experiments, and architectural direction.
Autonomy · continuity · evolution
- S25 The Claude 5 Family & Model Strategy Meet Claude Fable 5, Mythos 5, and the mid-2026 lineup — Opus 4.8, Sonnet 5, Haiku 4.5 — and how to pick and mix models per task in Claude Code.
- S26 Effort Levels, Ultrathink & Ultracode Choose Claude Code's reasoning effort deliberately, and understand the verified difference between ultrathink, ultracode, and enforceable CLI cost limits.
- S27 The Workflow Tool: Scripted Multi-Agent Orchestration Use Claude Code dynamic workflows to put repeatable JavaScript control flow around many subagents, with approval, progress, limits, and same-session resume.
- S28 Orchestration Quality Patterns Design multi-agent workflows around independent search, adversarial verification, explicit stopping, and evidence that remains auditable.
- S29 Persistent Memory: How Claude Code Remembers How Claude Code's file-based persistent memory works — one fact per file, a MEMORY.md index, four memory types, and hygiene rules that keep recall useful.
- S30 Deferred Tools & ToolSearch How Claude Code defers tool schemas and loads them on demand with ToolSearch — so large MCP tool surfaces stop taxing every context window.
- S31 Scheduled Autonomy: Loops, Wakeups & Cron Master Claude Code's scheduled autonomy: /loop dynamic mode, ScheduleWakeup timing, prompt-cache economics, and cron jobs for standing automation.
- S32 The New Tool Belt: Artifacts, Tasks & Typed Outputs A verified tour of Claude Code's current output and coordination tools, including live MCP-backed artifacts, task output files, LSP, and recent Week 28–29 changes.
No lessons match those filters.