코스
Claude 플랫폼 기초부터 각 시험 도메인까지 단계적으로 구성된 6개 코스입니다.
Claude Platform Foundations
Start here. Learn how the Claude Messages API actually works — roles, tool_use, streaming, prompt caching, token counting, and the Message Batches API — then where Claude excels and fails, and how to run your first Claude Code session.
Agentic Architecture & Orchestration
Design reliable agentic systems on Claude: master the agentic loop and stop_reason, decide when the model should drive control flow versus when to hard-code it, compose multi-agent topologies with isolated subagent context, configure agents and tool scopes with the Agent SDK, and manage sessions, decomposition, and human-in-the-loop escalation.
Tool Design & MCP Integration
Design tools Claude can select reliably, return structured errors, and integrate the Model Context Protocol end to end — from .mcp.json configuration and transports to Grep→Read→trace investigation with built-in tools.
Claude Code Configuration & Workflows
Configure and operate Claude Code like an architect: the CLAUDE.md memory hierarchy and @path imports, path-scoped .claude/rules/, custom slash commands vs Skills, plan mode and permission models, and headless CI/CD runs with structured output, /compact, /memory, and the Explore subagent.
Prompt Engineering & Structured Output
Write prompts Claude can act on reliably and produce machine-validatable output. Master explicit criteria, few-shot examples, chain-of-thought, and system/role prompts; then guarantee structured output with tool_use + JSON Schema, build validation-retry loops, and scale review with multi-pass instances and the Message Batches API.
Context Management & Reliability
Keep long-running Claude systems reliable. Engineer the context window against lost-in-the-middle and lossy summarization, propagate structured errors instead of swallowing them, annotate coverage and provenance, persist state across context boundaries, and optimize token budgets with prompt caching and stratified human review.