CERTIFICATION PREP

Pass the CCA-F exam on your first attempt

Five domain study guides, 39 scenario-based practice questions with full justifications, hands-on build exercises, and the exam strategy guide that tells you exactly what the exam rewards and penalises.

Get the Course

30–35 hours to exam readiness. Looking for the free version?

5
Domain study guides
39
Practice questions
720
Passing score (of 1000)
5
Build exercises

Everything between you and 720/1000

Not a crash course. A structured certification prep built around how the exam actually tests you — production scenarios, architectural tradeoffs, and the traps that fail most candidates.

📚

5 domain study guides

In-depth coverage of every exam domain, grounded in three real production scenarios: a customer support agent, a multi-agent research system, and developer productivity tools.

~20 hours of study material
🎯

39 practice questions

Scenario-based MCQs matching the real exam format. Every question includes the correct answer, full justification, and an explanation of why each distractor fails.

6–10 questions per domain
🛠

Build exercises

Hands-on implementation specs for each domain. Build a multi-agent research coordinator, design tool schemas, configure CLAUDE.md hierarchies. Each exercise includes a verification checklist.

~10 hours hands-on work
💡

5 interactive tutors

Teaching prompts for each domain that adapt to your skill level. They assess where you are, teach the gaps, then run a targeted practice exam with gap analysis.

Adaptive to your experience

Exam strategy guide

"Always choose" and "always reject" decision rules extracted from exam patterns. Know which answer the exam rewards before you sit down — not after you fail.

Cross-domain concept mapping
📑

Reference library

300+ Claude Code features catalog (every release from v0.2 to v2.1) plus 70+ curated links to official Anthropic documentation, organised by exam domain.

Always up to date

What the exam actually tests

Every question is a production scenario. You're the architect — what do you recommend? The weights tell you where to spend your study time.

27%
weight

Agentic Architecture & Orchestration

Loop lifecycle, multi-agent hub-and-spoke, subagent context isolation, workflow enforcement via hooks, task decomposition, and session state management.

Agentic loops Multi-agent patterns Hooks Session state Task decomposition
10 questions
20%
weight

Claude Code Configuration & Workflows

CLAUDE.md hierarchy (user vs project vs directory), custom commands and skills, path-specific rules, plan mode vs execution, and iterative refinement strategies.

CLAUDE.md Custom commands Skills Plan mode @import
8 questions
20%
weight

Prompt Engineering & Structured Output

Explicit categorical criteria, few-shot prompting, tool_use JSON schemas, validation-retry loops, and schema design that prevents fabrication.

Few-shot prompting tool_use schemas Validation loops Fabrication prevention
8 questions
18%
weight

Tool Design & MCP Integration

Tool descriptions as the selection mechanism, four error categories with retry guidance, tool_choice configuration, and build-vs-use decisions for MCP servers.

Tool descriptions Error categories MCP servers tool_choice
7 questions
15%
weight

Context Management & Reliability

Case facts preservation, escalation triggers (valid vs unreliable), error propagation patterns, context degradation after long sessions, and scratchpad strategies.

Context preservation Escalation Error propagation Scratchpad pattern
6 questions

Study, practice, build, pass

The course follows a deliberate progression. Each domain builds on the last. The exam strategy ties them together.

1

Study the domain guides

Work through each domain's study guide. Every concept is grounded in production scenarios — not abstract theory. Key exam traps are called out inline.

2

Take the practice exams

Each domain ends with 6–10 scenario-based questions. Read the justifications for every answer — including why the wrong ones are wrong. That's where the real learning happens.

3

Build to solidify

Build exercises turn theory into muscle memory. Implement a multi-agent coordinator, design tool schemas, wire up hooks. Use the verification checklist to confirm you got it right.

The decision rules that pass candidates use

The exam rewards specific architectural instincts. These patterns appear across all 5 domains.

The exam always rewards

  • stop_reason over natural language signals for loop termination
  • Programmatic enforcement (hooks, gates) for financial, security, and compliance
  • Better tool descriptions as the first fix for misrouting
  • Community MCP servers before building custom ones
  • Fresh start + summary injection over stale resumed context
  • Concrete examples over prose descriptions
  • Independent review instances (not self-review)
  • Nullable fields over required fields to prevent fabrication

The exam always penalises

  • Prompt-based solutions for high-stakes scenarios
  • Arbitrary iteration caps as the primary termination mechanism
  • Sentiment-based escalation triggers
  • Batch API for blocking workflows
  • Single-pass reviews of 10+ files
  • Silent error suppression or workflow termination on partial failure
  • Subagents inheriting full coordinator history
  • Self-reported confidence as an escalation signal

This is what the exam looks like

Every practice question follows this format: a production scenario, four plausible answers, and a detailed breakdown of why one is correct.

Domain 1 — Agentic Architecture
A multi-agent research system has a coordinator that spawns three specialist subagents (market analysis, competitive intelligence, technical assessment). The coordinator passes its full conversation history to each subagent for context. Reports are coming back with duplicated findings and the system is running 3x slower than expected.
What is the most likely root cause?
  • A. The subagents need higher token limits to process the full history
  • B. Subagents should receive only their specific task context, not the coordinator's full history
  • C. The coordinator needs a deduplication step after collecting reports
  • D. The system should run subagents sequentially instead of in parallel
Why B is correct: Subagents in a hub-and-spoke pattern should be isolated. Passing the coordinator's full conversation history causes attention dilution — subagents process irrelevant context, leading to duplicated findings and slower execution. The fix is at the input, not the output.

Why C is tempting but wrong: Deduplication treats the symptom (duplicated findings) but not the cause (context leakage). The 3x slowdown would persist.

Already using the free course?

It's a great start. Here's what CCA-F Pro adds to get you across the 720 line.

The free course gives you the foundation

32 interactive lessons and 109 practice questions inside Claude Code. You learn the concepts, build muscle memory with the tool, and test yourself with a mock exam. Start the free course →

CCA-F Pro adds
📚
5 domain study guides

~20 hours of structured material. Every exam domain covered in depth with production scenarios, not just definitions.

🔍
Full answer justifications

39 exam-format questions where every answer option is explained — why the right answer wins and why each distractor fails.

🛠
5 build exercises

Hands-on specs with verification checklists. You build real systems, not just read about them.

🎯
Exam strategy guide

Always-choose and always-reject decision rules. Know which patterns the exam rewards before you sit down.

🤖
Adaptive teaching prompts

5 interactive Claude Code prompts that adjust to your level. Drill weak areas until they're strengths.

🗺
Cross-domain mapping + reference library

Concepts that appear across multiple domains flagged, plus a 300+ feature catalog and 70+ official doc links.

CCA-F Pro

Complete Certification Prep

Everything you need to pass the Claude Certified Architect exam on your first attempt.

$49 one-time

Instant access. No subscription. Yours forever.

  • 5 domain study guides (~20 hours of material)
  • 39 scenario-based practice questions with full justifications
  • 5 build exercises with verification checklists
  • 5 interactive teaching prompts (adaptive to your level)
  • Exam strategy guide with decision rules
  • 300+ Claude Code features reference
  • 70+ curated official documentation links
  • Cross-domain concept mapping
Get CCA-F Pro — $49

Not satisfied? Email us within 7 days for a full refund.

Common questions

The free course runs inside Claude Code as 32 interactive lessons with 109 practice questions. It teaches the concepts. CCA-F Pro is structured around the exam itself — five comprehensive domain guides, practice questions that match the exam format with full justifications for every answer, hands-on build exercises, and an exam strategy guide. The free course teaches you Claude architecture. CCA-F Pro prepares you to pass the certification.
You can study the course regardless of your organisation. However, to register for the actual CCA-F exam, your company must be part of the Anthropic Partner Network. Current anchor partners include Accenture, Cognizant, Deloitte, and Infosys. Check with your company's Anthropic relationship owner to confirm eligibility before beginning your study.
Approximately 30–35 hours total: ~20 hours for the domain study guides, ~5 hours for practice exams and review, and ~10 hours for the build exercises. Most candidates spread this over 2–4 weeks.
The course is a downloadable package containing Markdown study guides, interactive teaching prompts you run inside Claude Code, and build exercise specifications. No videos, no LMS, no login. Study guides work in any Markdown reader. Teaching prompts run in Claude Code with a Pro or Max subscription.
No. This is an independent exam preparation resource built by The Crux AI. It is not affiliated with, endorsed by, or officially associated with Anthropic. "Claude" and "Claude Code" are trademarks of Anthropic, PBC.
Email us within 7 days of purchase for a full refund, no questions asked. hello@thecrux.ai
Yes, and we recommend it. Use the free course for interactive concept learning inside Claude Code, then use CCA-F Pro's study guides and practice exams for structured exam preparation. The adaptive teaching prompts in CCA-F Pro also run inside Claude Code, so you get the best of both approaches.