Claude Developer Foundations (CCDV-F) Study Guide

Domain 1: Agents and Workflows (14.7%)

Agent Architecture (4.5%). Principles, patterns and trade-offs of agent and workflow architecture; Deciding between a workflow and an agent; Manager/supervisor hierarchies; Role of subagents in task execution.

Agent Construction with Claude (5.3%). Claude Agent SDK; Custom agent loops and harnesses; Self-hosted vs Anthropic-hosted managed agent deployment; Hooks for deterministic actions.

Agent Patterns and Frameworks (4.9%). Tool-use loops, subagents, memory and context-window management; Agentic frameworks such as Strands, LangGraph and PydanticAI.

Study note: the recurring decision is workflow versus agent. Workflows for predictable, auditable steps; agents when the path depends on what is discovered. Know what the Agent SDK gives you (loop, tools, hooks, subagents) and what hosted versus self-hosted deployment changes.

Domain 2: Applications and Integration (33.1%)

Understanding Requirements (3.4%). Functional and infrastructure requirements derived from business requirements and solution architecture.

Systems Life Cycle (2.8%). Concepts and frameworks for developing, implementing, operating and maintaining IT systems.

Claude API Mechanics (6.8%). Messages, tools, streaming, vision, thinking, caching; Invoking Claude through third-party vendors; Messages API data access patterns; Batch API use; realtime vs batch trade-offs.

Software Engineering Foundations (7.4%). REST APIs, JSON, asynchronous programming, version control; SDLC integration, code review, small- and large-scale refactoring.

Claude Application Design (8.6%). How Claude interprets instructions across interfaces (Claude Code, Desktop, claude.ai, API, SDKs); Content boundaries, schema design, session hygiene, plugin management.

Configuration Management (4.1%). CLAUDE.md files, settings.json, model version pinning, prompt versioning, plugin dependencies.

Study note: know the Messages API surface cold — tools, streaming, vision, extended thinking, prompt caching — and when the Batch API beats realtime. Expect items that start from a business requirement and ask for the infrastructure consequence. Configuration management is about reproducibility: pinned model versions, versioned prompts, CLAUDE.md and settings.json under source control.

Domain 3: Claude Code (3.1%)

Claude Code Operation (3.1%). Core components (Rules, Skills, Commands, Agents, Agent Memory); Session management, built-in and custom slash commands, headless mode, streaming mode, auto-mode; CLAUDE.md hierarchy, repository initialisation, settings.json configuration.

Domain 4: Eval, Testing, and Debugging (2.6%)

Debugging and Error Handling (2.6%). Identifying error types and selecting recovery strategies; Trace analysis to find failure modes; Isolating problem origin between the integration layer and model output.

Domain 5: Model Selection and Optimization (16.8%)

LLM Fundamentals (5.2%). Tokens, context windows, sampling, non-determinism, next-token generation; Model options (fast mode, extended thinking, adaptive thinking, effort levels); Zero-shot, single-shot and multi-shot prompting.

Technical Fundamentals (6.1%). Integrating with SDKs that wrap REST APIs; Websockets.

Model Selection and Tradeoffs (2.7%). Opus vs Sonnet vs Haiku use cases; Adaptive thinking support; Quality/latency/cost trade-offs; Breaking behaviour changes across model releases.

Cost and Token Management (2.8%). Token budgeting, cost modelling, usage tracking; Prompt caching and cache check-pointing.

Study note: be able to explain tokens, context windows and non-determinism to a colleague, choose between Opus, Sonnet and Haiku on quality, latency and cost, and compute a cost estimate from token counts. Prompt caching and cache checkpoints are the most testable optimisation.

Domain 6: Prompt and Context Engineering (11%)

Context Engineering (3.8%). Context window management; Preventing context drift and bloat (tool output pruning, compaction); Context isolation through subagents or multi-step workflows.

Prompt Engineering (4.6%). Instruction clarity, few-shot examples, system vs user placement, output constraints; Prompt and instruction placement across components; Iterative refinement, prompt adjustment, input sanitisation.

Output Handling (2.6%). Structured output patterns, response validation, defensive parsing; Scepticism toward confident output.

Domain 7: Security and Safety (8.1%)

AI Application Security (3.2%). Data privacy and security; prompt injection awareness and mitigation; jailbreak defence; Untrusted input handling, data leakage prevention, PII handling; Authentication, authorisation, confidentiality, privacy and integrity.

Guardrails and Safe Deployment (2.3%). Content policy, guardrail layering; Secure-by-design principles (privacy, IAM, least privilege).

Claude Hooks (1%). Using hooks for guardrails and safety controls that prevent destructive actions.

Identity, Secrets, and Key Management (1.6%). Managing secrets, credentials and API keys across development and production; Identity validation, access approval and level verification, authorised access monitoring.

Study note: questions ask for the control, not the exploit: treat retrieved content as data, scope tool permissions, keep secrets out of prompts and repositories, layer guardrails, use hooks to block destructive actions.

Domain 8: Tools and MCPs (10.6%)

Tool Implementation (4.4%). Tool use and function calling; external system configuration; tool description writing; error handling; Usage patterns (agentic harness dispatch, client-side vs server-side tools, approval patterns); tool set construction.

MCP Server Development (2.1%). Server authoring, deployment and integration; MCP resources, tools and prompts; Communication patterns (stdio, sockets, client vs server).

Agentic Customization (4.1%). Trade-offs among built-in tools, custom tools, Skills and MCPs.

Study note: tool descriptions are the interface; vague descriptions cause wrong calls. Know MCP's three primitives (resources, tools, prompts) and the stdio versus socket transports, and when a Skill is the better abstraction than a tool or an MCP.