← 학습 가이드 목록으로 돌아가기
12주 학습 계획
완전 초보에서 시험 응시 수준까지 주차별 학습 계획을 안내합니다.
A focused 12-week path from zero to exam-ready for the Claude Certified Architect — Foundations (CCA-F) exam. Each week pairs a course module with deliberate practice. Aim for ~4–6 hours/week.
Weeks 1–2 — Platform foundations
- Course: Claude Platform Foundations (Module 1 The Claude API, Module 2 Capabilities & Limits).
- Learn the Messages API,
tool_useblocks, streaming, prompt caching, and the Message Batches API. - Practice: take the free diagnostic to find your weak domains.
Weeks 3–5 — Agentic Architecture (Domain 1, 27%)
- Course: Agentic Architecture & Orchestration (all four modules).
- Master the agentic loop and
stop_reason, hub-and-spoke multi-agent topologies, subagent isolation, the Agent SDK, sessions, and decomposition. - Practice: the Domain 1 quiz; review every explanation.
Week 6 — Tools & MCP (Domain 2, 18%)
- Course: Tool Design & MCP Integration.
- Tool descriptions as the selection mechanism,
tool_choice, structured errors, MCP fundamentals, config, and transports. - Practice: the Domain 2 quiz.
Weeks 7–8 — Claude Code Workflows (Domain 3, 20%)
- Course: Claude Code Configuration & Workflows.
- CLAUDE.md hierarchy, slash commands vs Skills, permission models, headless CI/CD.
- Practice: the Domain 3 quiz.
Weeks 9–10 — Prompt Engineering (Domain 4, 20%)
- Course: Prompt Engineering & Structured Output.
- Few-shot, chain-of-thought, JSON-Schema tool use, validation-retry loops, review patterns, Message Batches.
- Practice: the Domain 4 quiz.
Week 11 — Context & Reliability (Domain 5, 15%)
- Course: Context Management & Reliability.
- Lost-in-the-middle, provenance, scratchpads, token budgets, prompt caching.
- Practice: the Domain 5 quiz.
Week 12 — Full mock exams & review
- Read all eight scenario walkthroughs.
- Take a full-length mock exam under timed conditions. Review the per-domain breakdown.
- Re-take quizzes for any domain below ~70%. Skim the anti-patterns cheatsheet and flashcards the day before.
Exam day: 60 scenario questions, 120 minutes, scaled 100–1000, pass at 720. Flag-and-review; don't leave questions blank (no guessing penalty).