Claude Code Workflows: CCDV-F Developer Exam Guide

Claude code workflows are the layered configuration patterns through which developers encode persistent instructions, automate side effects, and structure multi-agent handoffs in Claude Code. For the CCDV-F exam, they surface across at least four of the eight domains, making workflow literacy a cross-cutting skill rather than a Domain 3 niche. Developers who study only the 3.1% Claude Code domain for workflow content miss the larger surface area in Domain 1 and Domain 2.

What does the CCDV-F exam cover, and where do workflows fit?

The Claude Certified Developer, Foundations exam (code: CCDV-F) has 53 items and a 120-minute time limit. The passing score is 720 on a 100-to-1,000 scale. Eight domains cover the full spectrum from API integration to security, with exact fractional weights published in the official exam guide:

Domain 2 alone accounts for roughly one-third of the exam. Claude Code workflow knowledge flows directly into Domain 2 scenario items about API configuration and runtime integration, so the domains are not siloed. Unlike the CCAR-F Architect exam, which draws 4 scenarios at random from a bank of 6 per sitting, the CCDV-F has no scenario bank: its 53 items are written directly against domain skills, so breadth of knowledge matters more than recognising specific scenario archetypes.

Claude Code Configuration & Workflows is the domain that most directly names these mechanics, but the underlying concepts recur in Domain 1 multi-agent orchestration items and Domain 8 tool-scoping scenarios as well.

What is CLAUDE.md and how does it shape a workflow?

CLAUDE.md is the primary mechanism for encoding persistent instructions into Claude Code. A file placed at the project root loads automatically at every session start and acts as a declarative baseline that shapes all downstream behaviour without any additional runtime configuration step.

Per Anthropic's Claude Code documentation, the configuration model follows a three-level configuration hierarchy: global user settings apply across all projects, a project-root CLAUDE.md applies to the entire repository, and subdirectory CLAUDE.md files apply only within their scope. The most narrowly scoped instruction wins when rules conflict.

This hierarchy has version control implications that the exam tests directly. Committing CLAUDE.md files to source control means every new team member inherits the full project context on first clone. A global CLAUDE.md might encode coding style and linting conventions; a subdirectory CLAUDE.md for a payments module might restrict available tools to read-only operations by default. When scopes conflict, the subdirectory rule takes precedence.

CCDV-F Domain 3 items about CLAUDE.md cluster around three questions: which configuration level takes precedence in a given scenario, what belongs in CLAUDE.md versus a slash command, and how @path imports affect scope when multiple files compose the effective configuration. Getting those three questions right covers most of Domain 3's workflow content.

For teams running Claude Code in CI pipelines, CLAUDE.md provides a stable anchor. A workflow that produces consistent output locally but behaves differently in headless CI mode usually has a scoping problem: either the CI environment is reading a different configuration level, or a tool restriction that applies locally is absent from the CI path.

How do hooks automate claude code workflows?

Hooks are shell commands that Claude Code executes at defined lifecycle points: before a tool is called (PreToolUse), after a tool completes (PostToolUse), at session start, and at session stop. They provide deterministic control over side effects, firing reliably regardless of what the model decides in any given response.

PostToolUse Hooks for Data Normalisation is one of the most exam-relevant patterns: a PostToolUse hook that reformats database query output after every read operation ensures that downstream prompts receive consistently structured data regardless of which developer is running the session. This is enforcement that a prompt instruction cannot reliably provide.

The Hooks vs Prompts Decision Framework supplies the exam-ready decision rule. Use a hook when the behaviour must fire regardless of the model's response and requires no reasoning. Use a prompt instruction when the model needs context to decide how to act. Masking a card number in every API response is a hook job; deciding whether an ambiguous escalation warrants human review is a prompt job.

Hooks also matter for security. A PreToolUse hook can intercept file-write operations and block any path that matches a protected pattern, providing a defence-in-depth layer that does not rely on the model correctly interpreting a prompt-level restriction. Domain 7 (Security and Safety, 8.1%) items about Claude Code almost always involve a hook-level control as the correct answer.

Domain 1 scenario items frequently present a multi-agent pipeline where a normalisation or validation step fires inconsistently because it was implemented as a prompt instruction. Identifying that pattern and specifying a PostToolUse hook as the proportionate fix is a recurring exam skill.

How do slash commands and skills extend claude code workflows?

Slash commands are on-demand shortcuts that invoke pre-written prompt templates or scripts. A /review-pr command might load a multi-step code-review workflow, run it against the current diff, and emit structured JSON. Commands are imperative and ephemeral: they fire when explicitly called and have no effect on sessions where they are not invoked.

The distinction between commands and CLAUDE.md is a reliable exam theme. CLAUDE.md instructions are declarative and always-on; slash commands are triggered and scoped to a single invocation. Mixing the two produces fragile workflows: a formatting rule encoded as a command fires only when the developer remembers to call it, whereas the same rule in CLAUDE.md fires on every session.

Skills extend commands with structured frontmatter that declares triggers, permitted tools, and model preferences. A skill file stored under .claude/agents/ can be invoked by name, delegating a well-scoped task to a subagent without polluting the main session's context. This is the design that Subagent Context Isolation describes: each subagent receives only the context it needs, keeping token usage predictable and preventing attention dilution in long-running workflows.

For the exam, the key judgment call is whether a subtask belongs in a skill or the main session. Skills are appropriate when the task is well-defined, the required tools are a strict subset of the parent's tools, and the output is verifiable before the parent continues. When those conditions do not hold, delegating adds coordination overhead without reliability gains.

What do CCDV-F exam items actually test about claude code workflows?

Domain 1 and Domain 3 items share a common structure: a realistic configuration is described, a failure or ambiguity is introduced, and you are asked to identify the root cause and the proportionate fix. Three failure patterns appear most frequently in practice:

1. Configuration scoping errors. A rule intended for a subdirectory was placed at the global level and now fires where it should not. The fix is to move the instruction to the appropriate CLAUDE.md scope, not to add a compensating rule elsewhere.
2. Hook vs prompt confusion. A behaviour that must fire reliably was implemented as a prompt instruction, making it probabilistic. The fix is to convert it to a hook at the correct lifecycle point.
3. Context leakage in subagent handoffs. A subagent receives the full parent session history when it should receive only a structured summary. The fix is structured context passing, not a larger context window.

The exam consistently rewards proportionate fixes. If a scoping error affects one subdirectory, the correct answer tightens that subdirectory's configuration rather than redesigning the entire hierarchy. This root-cause discipline carries through Domain 1 and Domain 2, where the same approach applies to API error handling and multi-agent recovery scenarios.

How should study time be allocated across workflow-related domains?

Allocate study time in proportion to domain weight, with an additional buffer on Domain 2 because its content overlaps every other domain. Based on the published weights applied to 53 items, Domain 2 contributes approximately 17 to 18 questions, Domain 1 approximately 7 to 8, and Domain 3 approximately 1 to 2.

Our adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold. When you answer Domain 3 items correctly in practice, the engine reallocates study time toward whichever domain has the largest remaining mastery gap, making preparation efficient rather than uniform across all eight domains.

The Agents and Workflows concept library covers the multi-agent patterns that underpin Domain 1. The intersection of those patterns with Claude Code's configuration primitives is exactly where the CCDV-F's harder workflow items live. Developers who master that intersection tend to find Domain 2's API integration scenarios more tractable, because the same configuration-scoping logic applies at the API level.

AI Skill Certs is an independent platform and is not affiliated with, endorsed by, or approved by Anthropic.