CertSafari

    Free Practice Questions for Anthropic Claude Certified Developer – Foundations (CCDV-F) - December 7 onwards Certification

    Exam guide version:
    1.1 (Effective December 7, 2026)
    Guide checked for updates:
    20 Nov 2026
    Question bank created:
    18 Nov 2026
    Question bank last updated:
    19 Nov 2026

    Study with 448 exam-style practice questions designed to help you prepare for the Anthropic Claude Certified Developer – Foundations (CCDV-F) - December 7 onwards. All questions are aligned with the latest exam guide and include detailed explanations to help you master the material.

    Your progress

    Coverage
    Mastery
    Performance
    Lessons

    LessonsNew

    24 documentation-grounded lessons, one per exam-guide subdomain — every claim cited to the official docs.Based on the official docs as of 18 Nov 2026

    Start Practicing

    Start a quiz

    Practice with randomly mixed questions from all topics

    Question MixAll Topics
    FormatRandom Order

    Exam experiences

    Pass and fail outcomes from candidates who prepared here — advice, scores, and prep time.

    Your saved questions

    Open the list of questions you bookmarked during practice for this exam.

    Study notes

    Private notes per question, grouped by exam domain — opens on its own page, not inline on this overview.

    Quiz History

    Exam Details

    Key information about Anthropic Claude Certified Developer – Foundations (CCDV-F) - December 7 onwards

    Official study guide

    View

    Question formats CertSafari offers
    • Multiple choice
    renewal:

    Free, non-proctored renewal assessment on the Anthropic Partner Academy

    exam code:

    CCDVF

    exam format:

    Scenario-based multiple choice; each item states how many responses to select

    registration:

    Anthropic Partner Academy, scheduled through Pearson

    guide version:

    1.1 (effective December 7, 2026)

    passing score:

    720 on a scaled range of 100–1,000

    prerequisites:

    None

    retake policy:

    Waits of 14, 30, then 90 days; max 4 attempts per rolling 12 months

    delivery method:

    Proctored via Pearson: OnVUE online or test center

    target audience:

    Software engineers, AI/ML engineers, Full-stack and backend developers, Solutions engineers

    registration fee:

    $125 USD

    result reporting:

    Pass/fail with scaled score and percent-correct by domain

    time limit minutes:

    120

    number of questions:

    70 (60 scored, 10 unscored)

    certification validity:

    12 months

    recommended experience:

    1–5 years of software engineering and 6+ months hands-on with Claude

    Exam Topics & Skills Assessed

    Skills measured (from the official study guide)

    1: Agents and Workflows

    1.1: Architecture and Runtime Selection

    Evaluating whether a workflow or an agentic architecture fits a set of business requirements, error tolerance, and audit constraints. This includes choosing the orchestrator and subagent decomposition and selecting the execution environment: a self-hosted harness or a managed, sandboxed agent runtime.

    1.2: Agent SDK Construction and Hooks

    Building and correcting Claude Agent SDK agent definitions, custom agent loops and harnesses, and hook configurations so that required actions execute deterministically rather than at the model’s discretion.

    1.3: Agent Pattern Diagnosis

    Analyzing a trace or description of a misbehaving agent to identify which design pattern is missing or misapplied, such as tool-use loop termination, subagent isolation, memory, or context-window management, and selecting the pattern that corrects the behavior.

    2: Applications and Integration

    2.1: Messages API Mechanics

    Applying Messages API mechanics in application code and API requests, including message structure, tool use, streaming, vision, thinking, prompt caching, structured output, and stop reasons, to produce the intended behavior, and choosing between realtime and batch processing.

    2.2: Application Boundary and Surface Behavior

    Evaluating an application design for how Claude interprets instructions on different surfaces, what content crosses the application boundary, and whether its schema and session decisions meet the stated requirements.

    2.3: Configuration and SDLC Integration

    Determining the correct configuration for a Claude system from its configuration artifacts and SDLC context, including CLAUDE.md files, settings.json, model version pinning, prompt versioning, plugin dependencies, and CI integration, and assessing how a configuration change affects each environment.

    2.4: Integration Resilience

    Analyzing failing or degraded integration code and its error output to find the defect and select a correction consistent with documented API error semantics. Defects include mishandled streaming events and terminal-event semantics, retry and fallback logic that does not distinguish Anthropic’s error classes (rate limits, overloaded, invalid request, timeout), and faulty multi-format ingestion through the Files API.

    3: Claude Code

    3.1: Claude Code Operation

    Operating Claude Code correctly across the CLI, desktop, IDE, web, and mobile surfaces, including session management, the CLAUDE.md hierarchy and precedence, settings.json, permission modes, headless and streaming modes, and built-in commands.

    3.2: Claude Code Customization and Extension

    Analyzing a repeatable team workflow to choose the extension mechanism that fits it, whether a Skill, plugin, Rules file, hook, subagent, or slash command, and authoring and packaging that extension for distribution.

    4: Eval, Testing, and Debugging

    4.1: Evaluation Design

    Constructing an evaluation that measures a stated quality goal for a Claude application or component, including LLM-as-judge harnesses calibrated against human-labeled data, golden sets, regression tests, and behavioral checks, and selecting defensible metrics.

    4.2: Failure-Mode Diagnosis

    Analyzing a trace excerpt or reproduction to identify the failure mode, such as hallucination, refusal, under-acting, context loss, format drift, or inconsistent output; isolating whether it originates in the integration layer or in model output; and selecting a recovery that addresses the cause rather than the symptom.

    5: Model Selection and Optimization

    5.1: Generation Mechanics and Model Options

    Applying an understanding of tokens, context windows, sampling and non-determinism, and model options such as adaptive thinking and effort levels to predict, explain, and control an observed behavior.

    5.2: Model Tier Selection and Release Planning

    Evaluating model tier options across quality, latency, and cost against task requirements and binding constraints, justifying the selection, and planning for behavior change across model releases using regression evaluations.

    5.3: Cost Diagnosis and Optimization

    Analyzing usage or cost data to identify the dominant cost driver, selecting the optimization that addresses it, such as prompt caching (single, layered, and rolling breakpoints and checkpoint placement), batch processing, or token budgets, and quantifying the expected impact.

    6: Prompt and Context Engineering

    6.1: Context Management for Long-Running Sessions

    Analyzing context bloat or drift in a long-running session or agent and selecting the management technique that restores intended behavior, such as compaction, tool-output pruning, subagent isolation, or criteria for starting a fresh session.

    6.2: Prompt Revision

    Revising a prompt to close a gap between intended and actual output that regression evaluations have exposed, using instruction clarity, system-versus-user placement, zero-, single-, or multi-shot examples, output constraints, and input sanitization, closing the gap and keeping it closed.

    6.3: Structured Output and Defensive Parsing

    Applying structured outputs with strict schema validation, response validation, and defensive parsing of free-text fallbacks to protect downstream code from malformed or unexpected model output.

    6.4: Grounding and Source Citation

    Constructing a request that grounds answers in the supplied documents, cites their source, and returns ‘not found’ when the documents do not contain the answer. This includes deciding, based on corpus size, cost, and freshness, whether to supply the full document set in context or retrieve only the relevant passages.

    7: Security and Safety

    7.1: Prompt Injection and Jailbreak Defense

    Analyzing the prompt-injection and jailbreak exposure of an agent that processes untrusted content, and selecting layered mitigations that prevent data leakage and unintended actions without breaking the agent’s function.

    7.2: Deterministic Guardrails and Least Privilege

    Applying guardrail layering, least privilege, and hooks as deterministic enforcement so that destructive actions cannot execute even under adversarial instruction.

    7.3: API Key and Credential Handling

    Handling Anthropic API keys and credentials correctly in development and production, including server-side proxying rather than client embedding, workspace-scoped keys versus admin keys, environment separation, rotation, and access monitoring, without exposing any key beyond its intended scope.

    8: Tools and MCP

    8.1: Tool Design and Response Shaping

    Implementing a tool for an integration need: a function schema, a description that drives correct invocation, response shaping so the model receives a minimal and actionable result, error handling, the choice of execution side, and an approval pattern, and assembling the tool set for performance.

    8.2: MCP Server Development and Integration

    Building, deploying, and integrating an MCP server into Claude applications and workflows, including its resources, tools, and prompts, the choice of transport (stdio or HTTP), and authentication.

    8.3: Capability Mechanism Selection

    Evaluating built-in tools, custom tools, Skills, and MCP servers for a use case, based on where the capability must live, who maintains it, and how Claude discovers and invokes it, and justifying the selection.

    Techniques & products

    Claude Developer Platform
    Messages API
    Anthropic SDKs (Python, TypeScript)
    Claude Agent SDK
    Claude Managed Agents
    Claude Code
    Claude Code CLI
    Claude Code desktop, IDE, web and mobile
    Model Context Protocol (MCP)
    MCP servers
    Files API
    Message Batches / batch processing
    Prompt caching
    Streaming events
    Stop reasons
    Tool use
    Vision
    Extended and adaptive thinking
    Effort levels
    Structured outputs
    JSON schema validation
    CLAUDE.md
    settings.json
    Permission modes
    Headless mode
    Skills
    Plugins
    Rules files
    Hooks
    Subagents
    Slash commands
    Workflow vs agentic architectures
    Orchestrator-subagent patterns
    Agent harnesses
    Context compaction
    Context-window management
    LLM-as-judge evaluation
    Golden sets
    Regression evaluations
    Model tiers
    Model version pinning
    Prompt versioning
    Token budgets
    Retrieval and document grounding
    Citations
    Prompt injection defense
    Jailbreak defense
    Least privilege
    Deterministic guardrails
    API key management
    Workspace-scoped and admin keys
    CI/CD integration
    Git
    GitHub Actions
    Python
    TypeScript
    REST APIs
    JSON

    CertSafari is not affiliated with, endorsed by, or officially connected to Anthropic PBC. Full disclaimer