CCA-F Study Guide — How I’m Preparing to Score 85%+ on Claude Certified Architect Foundations
288 questions. 17 domains. Visual flowcharts. Every answer explained. A complete study deck built from the official exam content — designed the way I actually learn.
Why CCA-F — and Why This Deck
Anthropic’s Claude Certified Architect — Foundations (CCA-F) exam tests whether you can design, build, and operate production Claude systems. It covers everything from agentic loop patterns and hook-based guardrails to MCP server architecture, extended thinking configuration, and prompt caching strategies.
Most certification prep material is walls of text. I don’t learn that way — and I’m guessing you don’t either. So I built a visual study system: a 614-slide deck that teaches each domain through diagrams and flowcharts first, then tests you with every exam question in a Question → Click → Answer + Explanation format.
The goal is 85%+. Here’s the system I’m using to get there.
8 Rules to Score 85%+
After going through all 288 questions, these are the patterns the exam hammers hardest. Memorize these and you’ll answer half the questions on autopilot:
Root Cause Thinking
Most questions test WHY something fails, not WHAT to do. Always look for the deepest cause in the architecture.
Hooks > Prompts
Hooks = 100% deterministic enforcement. Prompts ≈ 95%. For financial ops and compliance, the answer is ALWAYS hooks.
Isolated Subagent Contexts
Subagents do NOT inherit coordinator context. The coordinator must explicitly pass all data in each subagent’s prompt.
Decomposition = Coordinator
If subagents worked correctly but output has gaps → root cause is the coordinator’s task decomposition, not the subagents.
Fresh > Resume
When code changes outside a session, start fresh with a structured summary. Stale tool results = incorrect recommendations.
budget_tokens = Maximum
Extended thinking budget is a ceiling, not guaranteed usage. Min 1,024. Incompatible with forced tool_choice. Temp must be 1.0.
Cache Order Matters
Static content FIRST → cached docs → dynamic query LAST. Prefix must match exactly. Ephemeral TTL = 5 minutes.
Loop on tool_use Only
Agentic loop continues ONLY on stop_reason='tool_use'. Exit on end_turn, max_tokens, refusal, stop_sequence.
What’s in the Study Deck
Visual Concept Diagrams
Each major domain opens with architectural diagrams — not bullet points. These are the slides I built to actually understand the patterns before touching a single question:
Agentic Loop Flowchart
continue/exit decision tree with color-coded stop_reason paths
Coordinator-Subagent Architecture
delegation pattern with isolation callouts and context passing
Context Contamination vs Clean Review
side-by-side biased vs objective instance comparison
Hook Execution Flow
PreToolUse → allow/deny branching with exit codes and PostToolUse
MCP Architecture
client → transport (stdio/HTTP) → server → tools/resources/prompts
Prompt Caching Order
cascading width bars: static → cached → dynamic ordering
288 Q&A Slides — Answer on Click
Every exam question gets a two-slide treatment:
Read the scenario + 4 options (color-coded circles A/B/C/D). Think through your answer. A yellow
▶ CLICK NEXT prompt at the bottom.
Correct answer in a green callout. Wrong options struck through and dimmed. A
WHY THIS IS CORRECT box with detailed reasoning — not just “A is right” but why B, C, D are wrong.
Sample Question
A coordinator agent manages three research subagents investigating ‘AI applications in creative industries.’ All three subagents covered visual arts extensively but missed music, creative writing, and film entirely. The subagents executed their tasks correctly. What is the root cause?
Domain Breakdown — 17 Exam Areas
| Domain | Questions | Weight |
|---|---|---|
| Agentic Architecture & Design Patterns | 78 | 27% |
| Prompt Engineering & System Prompts | 40 | 14% |
| Tool Design & Tool Use | 35 | 12% |
| Streaming & API Mechanics | 22 | 8% |
| MCP (Model Context Protocol) | 18 | 6% |
| Context Window & Token Management | 17 | 6% |
| Hooks, Guardrails & Agent SDK | 15 | 5% |
| Extended Thinking & Reasoning | 14 | 5% |
| Multi-turn Conversations | 12 | 4% |
| Error Handling & Reliability | 10 | 3% |
| Caching & Performance | 7 | 2% |
| Evaluation & Testing | 6 | 2% |
| RAG & Knowledge Management | 6 | 2% |
| Cost Optimization & Batch Processing | 4 | 1% |
| Security, Safety & Trust | 2 | 1% |
| Model Selection & Capabilities | 1 | <1% |
| Vision & Multimodal | 1 | <1% |
5 Concepts That Appear Everywhere
PreToolUse hooks enforce compliance at 100% (exit code 0 = allow, exit code 2 = deny). System prompt instructions work ~95% of the time. For anything financial, security, or compliance — hooks, not prompts.
stop_reason='tool_use'. Exit on end_turn, max_tokens, refusal, or stop_sequence. This pattern appears in at least 15 questions.
📥 Download the Complete Study Deck
614 slides. 288 questions with answers. Visual diagrams for every major domain.
Dark theme. Brand-matched. Ready for PowerPoint, Keynote, or Google Slides.
How I’m Using This
My study loop is simple: I go domain by domain. First I review the concept diagram to build mental architecture. Then I click through the Q&A slides — I read the question, pick my answer mentally, click next, and check. When I get one wrong, I re-read the explanation and note the pattern.
After each domain, I do the cheat sheet review — 18 concepts on two slides, each one a single line I can recall under pressure. Then I move to the next domain.
The goal is three full passes through the deck before the exam. First pass to learn. Second pass to identify weak spots. Third pass the day before, focusing only on domains where I scored below 85%.
Share
If you’re also prepping for the CCA-F, grab the deck above and let me know how it goes. I’ll update this post with my score once I sit the exam. 🎯