Claude Code on CCDV-F: two items, one skill, and where the rest of the vocabulary went

Claude Code on CCDV-F is a single skill, Claude Code Operation, published at 3.1% of the exam, which a 53-item allocation turns into about two items. The skill is an inventory: five components, five features, the instruction-file hierarchy, repository initialisation and settings.json. The items ask which part of that inventory fits a stated need, and the reflex that loses the mark is choosing an instruction where the scenario needed a setting, or a faster mode where it needed a narrower one.

What Claude Code on CCDV-F actually lists

Source: the CCDV-F exam guide, version 1.0, effective July 2026, section 6, Domain 3; the two-item figure is a 53-item allocation of the published weight. Checked against the official sources on 17 September 2026.

CCDV-F is the only one of the four papers that publishes a weight for every skill, and here the domain and its skill are the same thing, so 3.1% is the whole story. On the Claude Certified Developer Foundations paper only Eval, Testing and Debugging is lighter, at 2.6%. The Claude Code overview is the official reference for the component list.

Where the rest of the vocabulary went

Search the CCDV-F blueprint for Claude Code, CLAUDE.md, settings.json, plugins or hooks and five skill descriptions match. Only one is Domain 3. Claude Application Design names Claude Code as one of the interfaces an application must reason about, and plugin management with it. Configuration Management names CLAUDE.md files and settings.json beside version pinning and plugin dependencies. Agent Construction names hooks for deterministic actions, and Claude Hooks is a skill of its own.

That distribution changes how to read a scenario. An item about pinning a model version or versioning a prompt is a Configuration Management item under Applications and Integration, asking the change-management question: what breaks when this moves. An item about a hook that must block a destructive action is a Security and Safety item. Domain 3 asks the operating question: which component, which feature, which level of the hierarchy.

Five components, five different jobs

The published list is Rules, Skills, Commands, Agents and Agent Memory, and an item in this domain describes a need and offers four of them. The decision rule asks two things: who triggers it, and where it lives. A rule is loaded by the paths it is scoped to. A skill is chosen by the model when a task matches its description. A command is invoked by a person, by name. An agent runs delegated work in its own context. Memory is what an agent keeps between sessions, in files rather than conversation.

The second half of the rule, where it lives, is the hierarchy. Instruction files and settings both run from a managed level the organisation controls, through user and project, to a local level that never leaves the machine, and the level a rule is written at decides whom it binds. How those levels combine is the subject of Claude Code configuration for the exams; the official references are the memory, settings and skills pages of the Claude Code documentation. Repository initialisation gives you a starting instruction file, not a finished configuration, which is its own small trap.

Modes are about who is watching

The five features in the skill sort by one question: is there a person at the keyboard, and who reads what comes out. Session management is picking up earlier work rather than re-explaining it. Slash commands, built in or custom, are a person invoking a procedure. Headless mode is a run with nobody present. Streaming mode is output emitted as it is produced for a program to consume. Auto-mode approves actions that pass an automated check against what you asked for, and the explicit rules you wrote still apply.

Two decision rules follow. In a headless run, anything that would stop to ask a question has to be decided before the run starts, because a prompt with nobody to answer it is a failure, not a pause. And fewer prompts is not fewer safeguards: the mode that answers approvals for you and the mode that removes the checks are different things, and offering the second where the scenario wanted the first is a recognisable distractor. The permissions page is the official reference.

A worked CCDV-F Domain 3 item

Two items on the free timed CCDV-F mock, filtered to Domain 3, follow the same shape. That page reports a raw count; the exam reports a scaled score with a 720 cut, and no conversion is published.

The traps on Claude Code questions on CCDV-F

Five recur. Treating an instruction file as enforcement, when only a setting or a hook can guarantee anything. Removing the checks when the scenario asked to remove the prompts. Confusing a skill, which the model selects from its description, with a command, which you invoke. Assuming initialising a repository produced a complete configuration. And spending Domain 2 study time here because the tooling is familiar.

The last one is the expensive one. If you are deciding between tracks, the guide to choosing a Claude certification shows where the same tooling carries real weight: Claude Code Configuration and Workflows is 20% of CCAR-F, more than six times the developer share, and that is the paper for a candidate whose strength is this domain.

Key takeaways

• One skill, 3.1%, about two items. Only Eval, Testing and Debugging is lighter on the developer paper.
• The vocabulary is worth 19% elsewhere. Four other skills name the same tooling, and Configuration Management alone outweighs this domain.
• Pick a component by trigger and scope. Path-loaded rule, model-chosen skill, user-invoked command, isolated agent, file-based memory.
• Headless means decided in advance. A prompt with nobody to answer it is a failure, and the fix is a setting, never an instruction.
• Fewer prompts is not fewer safeguards. Answering approvals for you is not the same as removing the checks.
• Cap the time. Two marks here against seventeen in Applications and Integration.