Sixteen focused modules: from npm install through hooks, subagents, MCP servers, and the Agent SDK, plus the API foundations, prompt engineering, evaluation, RAG, and workflow patterns that make professional Claude Code work click.
For developers who want both depth and shipping speed: enough API knowledge to debug skill, subagent, and hook failures — without slogging through the full 30-module agent course.
Start with CC0 → or browse all 16 modules ↓The full 30-module course is a comprehensive agent-engineering curriculum. This track is purpose-built for Claude Code: every concept (API, prompts, RAG, evals, workflows) is anchored in skills, subagents, hooks, and MCP servers you'll actually ship.
| Topic | Direct Track (this course, 16 modules) | Full Agent Course (30 modules) |
|---|---|---|
| Audience | Developers using and extending Claude Code | Engineers building autonomous agents from scratch |
| Starting point | Install → first session in 10 min | Mental models, tokens, LLM internals |
| Code examples | Bash, JSON config, Python + TS for SDK use | Python + TypeScript, Anthropic SDK, deep deployment |
| Coverage | CLI features + API/prompts/eval/RAG/workflows tied to CC use cases | All of that plus M00 lifecycle, multi-agent systems, observability, capstones |
| End state | You ship production CC workflows, write your own MCP servers, and build standalone agents with the Agent SDK | You build production agents end-to-end |
# 1. Install (Node 18+) $ npm install -g @anthropic-ai/claude-code # 2. Authenticate (opens browser; use Pro / Team / Enterprise account or API key) $ claude # 3. Open it inside any repo $ cd ~/your-project && claude
Detailed setup — including IDE extensions, login modes, and the /help tour — is the first half of CC0.
Linear path. Each module assumes the previous. Foundations → CLI features → tools & integrations → production patterns → building your own agents.
/deploy, /review, /triage. Frontmatter, $ARGUMENTS, shell injection, bundled skills, context: fork.
CC6 · 50 min
Subagents
Built-in (Explore, Plan, general-purpose), custom YAML, fork mode, persistent agent-memory, isolation worktrees, the @agent- mention.
CC7 · 50 min
Hooks
All 28 lifecycle events, 5 handler types (command/http/mcp_tool/prompt/agent), async + defer permission, the /hooks menu.
Hierarchical CLAUDE.md, path-specific rules, project-shared .claude/ files, gitignored local overrides, AGENTS.md compatibility.
Permission modes, allow/deny rules, OS-level sandbox with filesystem & network walls, managed-policy settings for teams.
Custom skills with shell-injected context, frontmatter-controlled invocation, subagents that learn across sessions.
Lint-on-write, refund-limit guardrails, central policy via HTTP hooks, deferred permissions, pipeline chaining.
GitHub, Postgres, Slack via official servers, plus your own FastMCP servers exposing tools, resources, prompts.
Chunking, embeddings, BM25, hybrid retrieval — exposed as MCP tools so any Claude Code session benefits.
Test datasets, code & LLM-as-judge graders, regression gates in CI — make prompts refactor-safe.
Headless mode, two-Claude review, GitHub Actions, structured output via --output-format and --json-schema.
Same engine as Claude Code, exposed as a Python & TypeScript library. Define custom tools via in-process MCP, wire to existing APIs, ship standalone agents.
If you're here to build autonomous agents end-to-end — deployment, observability, multi-agent systems, capstone projects — the full 30-module course is the right home. This Direct Track shares the M25/M26 reference material; the full course adds 28 more modules around it covering architecture, tracing, monitoring, and three domain-anchored capstones.