Home › Certifications › Claude Code for Engineers
Claude Certification Guide · 2026Claude Code for Engineers: What the Claude Certification Exams Expect You to Know
Claude Code Configuration & Workflows carries 20% of the CCAR-F blueprint — joint second-largest domain on the paper — and it appears on CCDV-F as well. Using Claude Code daily is not the same as being examined on it. Here is the gap.
- Claude Code is 20% of CCAR-F, tied with Prompt Engineering & Structured Output and behind only Agentic Architecture & Orchestration at 27%.
- The exams test configuration and workflow design, not command recall.
- CI/CD integration is the most under-prepared area — most engineers use Claude Code interactively and have never put it in a pipeline.
- Claude Code and the Agent SDK are different tools for different jobs, and knowing when to reach for each is examinable.
- Permissions and boundaries matter — what an agent is allowed to touch is a design decision, not a setting.
- It is closed book. You cannot check a flag mid-exam.
Of the five CCAR-F domains, Claude Code is the one candidates most often assume they can skip preparing for. They use it every day, so how hard can a fifth of the paper be? Then the scenarios ask about pipeline integration, permission boundaries and team configuration — and daily interactive use turns out to have taught almost none of it.
Where it sits in the blueprint
| CCAR-F domain | Share |
|---|---|
| Agentic Architecture & Orchestration | 27% |
| Claude Code Configuration & Workflows | 20% |
| Prompt Engineering & Structured Output | 20% |
| Tool Design & MCP Integration | 18% |
| Context Management & Reliability | 15% |
Weightings from Anthropic's published CCAR-F exam guide via Anthropic Partner Academy. Anthropic revises these guides — the CCAR-F guide moved from v0.2 (30 June 2026) to v1.0 (July 2026) — so read the current version for your exam. Claude Code also appears within CCDV-F's coverage of engineering workflows. Checked 25 August 2026.
One fifth of the marks is roughly twelve questions on a 60-question paper. That is more than enough to decide a pass at 720/1000.
What the exams actually test
The distinction that matters: interactive use teaches you how to ask. The exams test how to configure, integrate and bound.
| Daily use teaches you | The exam asks about |
|---|---|
| Prompting Claude Code well in a session | Configuring it consistently for a whole team |
| Running it in your terminal | Running it unattended in a pipeline |
| Approving actions as they come up | Deciding in advance what may never be auto-approved |
| Working in one repository | Standardising across many |
| Fixing a failure when you see it | Designing so the failure class cannot occur |
Configuration
Project-level versus user-level configuration, what belongs in version control, and how a team keeps behaviour consistent between machines. Scenario questions typically present a team with inconsistent results and ask what should change — the answer is usually configuration discipline rather than better prompting.
CI/CD and unattended execution
This is the most under-prepared area, and the highest-yield thing to study.
Claude Code in a pipeline is a different problem from Claude Code at a prompt. There is nobody to approve an action, nobody to notice a loop, and failures are asynchronous. Expect questions on what is safe to automate, where a human gate is required, how failures surface, and how you keep a non-deterministic step from producing a non-reproducible build.
Put Claude Code into a pipeline once, on something trivial, and watch what happens when it needs a decision nobody is there to make. That hour teaches more about this domain than any amount of reading, and it maps directly onto how the scenarios are written.
Permissions and blast radius
What may Claude Code read, write, execute or delete? Under whose identity? Which actions require explicit human approval regardless of convenience?
The examinable insight is that these are architecture decisions, not preferences. "Allow everything and pay attention" is a design, and a bad one. Expect scenarios where the correct answer narrows what the agent can reach rather than adding a check after the fact.
Workflow design
Where Claude Code fits in a delivery process: which tasks suit it, which do not, how its output is reviewed, and how you keep review meaningful when volume rises. This overlaps with the orchestration domain, which is deliberate — the exams treat tooling and architecture as one subject.
Claude Code or the Agent SDK?
A genuine point of confusion, and examinable in both papers.
| Claude Code | Claude Agent SDK | |
|---|---|---|
| Is | An agentic coding tool you use | A library you build agents with |
| Audience | Engineers doing engineering work | Engineers shipping agents to users |
| You configure | Its behaviour and permissions | Its tools, loop and boundaries in code |
| Reach for it when | The agent's job is to help build software | The agent is part of the product |
| Weighted on | CCAR-F 20%; also CCDV-F | CCDV-F primarily |
The trap is treating them as competitors. They solve different problems, and a scenario naming the wrong one is often testing exactly that distinction. For the SDK side, see the CCDV-F prep course.
Review, and what happens when volume rises
This is the workflow question the exams keep circling, and it is worth understanding properly because it is genuinely unresolved in the industry.
An agentic coding tool changes the ratio between producing code and reviewing it. Producing gets dramatically cheaper; reviewing does not. Teams that adopt Claude Code without adjusting anything discover the bottleneck simply moved — they now have more change to review than the same number of people can meaningfully review.
Scenario questions in this area tend to reward answers that reduce what needs reviewing rather than answers that ask people to review faster:
- Narrower task scope so each change is small enough to actually read.
- Automated verification that catches whole classes of problem before a human looks — tests, types, linting, contract checks.
- Bounded permissions so the risky categories of change cannot be made unattended at all.
- Explicit review criteria, so reviewers know what they are checking for rather than skimming.
"The team should review carefully" is almost never the expected answer. It is a hope, not a design.
Configuration is a team problem, not a personal one
The most common real-world failure this domain examines is quiet configuration drift.
One engineer tunes their setup, gets good results, and shares neither the configuration nor the reasoning. A colleague on the same repository gets different behaviour and concludes the tool is unreliable. Nobody is wrong, and nothing is written down.
The examinable principle: configuration that affects output belongs in version control alongside the code it affects. If behaviour differs between two people on the same repository, that is a defect in the setup, not a quirk of the tool. Scenario answers consistently favour making behaviour explicit, shared and reviewable over making any individual's results better.
How to prepare for this domain
- Read the current exam guide first and confirm the weighting for your exam version.
- Move one repository to committed, shared configuration rather than per-machine settings.
- Run Claude Code unattended once, in a pipeline, on something low-stakes.
- Write down your permission boundaries explicitly — what may never be auto-approved, and why. Scenario answers follow this reasoning.
- Practise closed book. No flag-checking. If you cannot explain a configuration choice without the docs, you do not know it yet.
Claude Code ships quickly, and material more than a few months old often describes behaviour that has changed. Prefer Anthropic's current documentation over third-party tutorials, including this one, for anything version-specific. Where our guidance and the current docs disagree, the docs are right.
Claude Code and the Agent SDK, taught in exam proportion
Claude Code sits inside the CCDV-F blueprint alongside tool use, MCP servers, model selection and prompt caching. The 360DT Claude Certified Developer Prep — Foundations course covers all eight domains live, weighted the way the exam weights them, so you spend your time where the marks actually are.
Explore the CCDV-F Prep Course
Frequently asked questions
Is Claude Code on the certification exams?
Yes. Claude Code Configuration & Workflows carries 20% of the CCAR-F blueprint, making it joint second-largest behind Agentic Architecture & Orchestration at 27%. It also appears within CCDV-F's coverage of engineering workflows.
Is a Claude Code course enough to pass a Claude certification?
No. On CCAR-F it addresses 20% of the paper. Orchestration, prompt engineering and structured output, tool design and MCP, and context management make up the other 80%. Budget your study by blueprint weighting rather than by the topic you find most comfortable.
What is the difference between Claude Code and the Claude Agent SDK?
Claude Code is an agentic coding tool you use to do engineering work. The Claude Agent SDK is a library you build agents with, for agents that form part of a product. They solve different problems, and knowing which to reach for is itself examinable.
Do I need to memorise Claude Code commands and flags?
Not as trivia, but bear in mind the exam is closed book, so you cannot check anything. What is actually tested is configuration and workflow judgement: what to standardise, what to automate, what requires a human gate, and how to bound what the agent can reach.
Is there an official Anthropic Claude Code certification?
There is no standalone Claude Code credential. Claude Code is examined as a domain within the broader Claude Certified Architect and Developer certifications. Anthropic Academy also offers free self-paced Claude Code course material, which issues a certificate of completion rather than a certification.
Preparing for the architecture paper? See the CCAR-F prep course and how to pass CCAR-F. For the protocol domain, read MCP explained for certification candidates. Not sure which exam you need? Start with the Claude certification path for 2026.
About this guide. 360 Digital Transformation is an independent training provider. We are not Anthropic, we do not issue Claude credentials, and our courses are exam preparation rather than official Anthropic training. Claude Code changes frequently — verify version-specific behaviour against Anthropic's current documentation. Checked 25 August 2026.