Claude Certified Architect (CCA-F) - 6 Anthropic Style Exams

This practice test bundle contains multiple full-length practice exams for the Anthropic Claude Certified Architect — Foundations (CCA-F) exam. Each exam is built from a pool of original questions, weighted by domain to match the official exam blueprint.

D1 — Agentic Architecture & Orchestration (27% of the exam) — covers defining agentic systems — autonomy, tool use, and the action loop, agents vs. workflows vs. conversational systems, when to use agentic architecture vs. simpler patterns, task analysis and decomposition strategies, sequential vs. parallel execution patterns, dynamic planning, replanning, and ambiguity handling, orchestrator-subagent model — roles, scope, and context isolation, multi-agent topology patterns — hub-and-spoke, pipeline, peer-to-peer, agent communication, handoff schemas, and error propagation, state and session management — in-context vs. external memory, error classification — tool, reasoning, and environment errors, fallback and retry strategies, programmatic enforcement vs. prompt-based guardrails, human-in-the-loop escalation design. You will understand how each of these areas is tested on the exam and how they connect to real-world practice.

D2 — Claude Code Configuration & Workflows (20% of the exam) — covers claude code architecture — tool system and execution model, project setup, context management, and session continuity, user, project, and team-level claudemd — hierarchy and precedence, path-specific rules, subdirectory config, and claudeignore, writing effective claudemd instructions for teams, custom slash commands — creation, structure, and distribution, skills — skillmd, frontmatter, triggers, and enterprise deployment, subagents in claude code — configuration, scope, and delegation, hooks — lifecycle events, implementation, and exit code conventions, useful hook patterns — logging, safety nets, and automation, claude code sdk — programmatic session control, ci/cd integration — non-interactive mode and the -p flag. You will understand how each of these areas is tested on the exam and how they connect to real-world practice.

D3 — Prompt Engineering & Structured Output (20% of the exam) — covers clarity, specificity, and instruction design, system prompts — structure, role definition, and context injection, xml tags and document structure for complex prompts, few-shot prompting and high-quality example design, chain-of-thought and extended thinking, prefilling, output steering, and format anchoring, temperature, top_p, top_k — generation parameter control, multi-turn conversation design and context accumulation, json mode and schema-constrained output, tool use for structured data extraction and validation loops, prompt evaluation — test datasets, grading, and regression testing. You will understand how each of these areas is tested on the exam and how they connect to real-world practice.

D4 — Tool Design & MCP Integration (18% of the exam) — covers tool descriptions as routing mechanisms, tool input schema design — parameters, enums, and constraints, tool error handling, idempotency, and partial success, mcp architecture — servers, clients, and the three primitives, building mcp servers — defining and exposing tools, resources and prompt templates in mcp, mcp clients — discovery, invocation, and multi-server routing, mcp transport mechanisms — stdio vs. streamablehttp, advanced mcp features — sampling, notifications, and roots, mcp security — access scoping, authentication, and production hardening. You will understand how each of these areas is tested on the exam and how they connect to real-world practice.

D5 — Context Management & Reliability (15% of the exam) — covers context window architecture — tokens, limits, and cost tradeoffs, the lost in the middle effect — evidence and mitigation, rag architecture — chunking, embedding, and the retrieval pipeline, semantic search, bm25, and hybrid retrieval, multi-index rag, production hardening, and citation, prompt caching — eligibility rules, cache placement, and cost impact, batch api — 50% cost savings, 24-hour window, and workload design, long-conversation coherence, compaction, and session memory. You will understand how each of these areas is tested on the exam and how they connect to real-world practice.

Exams are arranged in graduated difficulty — from warm-up through exam simulation — so you can build confidence before testing under real conditions. Every question includes a detailed explanation covering why the correct answer is right and why each distractor is wrong.

Major topics covered: defining agentic systems — autonomy, tool use, and the action loop, agents vs. workflows vs. conversational systems, when to use agentic architecture vs. simpler patterns, task analysis and decomposition strategies, sequential vs. parallel execution patterns, dynamic planning, replanning, and ambiguity handling, orchestrator-subagent model — roles, scope, and context isolation, multi-agent topology patterns — hub-and-spoke, pipeline, peer-to-peer, agent communication, handoff schemas, and error propagation, state and session management — in-context vs. external memory, error classification — tool, reasoning, and environment errors, fallback and retry strategies, programmatic enforcement vs. prompt-based guardrails, human-in-the-loop escalation design, claude code architecture — tool system and execution model, project setup, context management, and session continuity, user, project, and team-level claudemd — hierarchy and precedence, path-specific rules, subdirectory config, and claudeignore, writing effective claudemd instructions for teams, custom slash commands — creation, structure, and distribution, skills — skillmd, frontmatter, triggers, and enterprise deployment, subagents in claude code — configuration, scope, and delegation, hooks — lifecycle events, implementation, and exit code conventions, useful hook patterns — logging, safety nets, and automation, claude code sdk — programmatic session control, ci/cd integration — non-interactive mode and the -p flag, clarity, specificity, and instruction design, system prompts — structure, role definition, and context injection, xml tags and document structure for complex prompts, few-shot prompting and high-quality example design, CCA-F exam prep 2026.