Practice Field Manual · Vol. 1

Claude Certified Architect
Foundations

Exam format · 4 of 6 scenarios
Passing score · 720 / 1000
Question bank · —
§ 00 — Orientation

A study companion built to mimic the real thing.

Every session randomly draws four of six scenarios, six questions per scenario, mirroring the Foundations exam structure. Answer choices are shuffled each time. Questions you've seen are remembered for the duration of this browser session — fresh material surfaces first until the pool runs out.

Scenarios
6total
Question Bank
213items
Per Session
24items
Pass Mark
720/1000
§ 01 — Exam Domains
I. Agentic Architecture 27%
II. Tool Design & MCP 18%
III. Claude Code Configuration 20%
IV. Prompt Engineering 20%
V. Context Management 15%
VI. Synthesis & Reliability Applied
§ 02 — Anti-Patterns Cheatsheet

The 18 most common wrong answers you'll encounter on the exam — memorize these to eliminate 2–3 options on sight before even reading the correct answer.

Domain 1 · Agentic Architecture
✗ Parsing natural language for loop termination CRITICAL
✓ Check stop_reason field (tool_use vs end_turn)
✗ Arbitrary iteration caps as primary stopping mechanism CRITICAL
✓ Let the agentic loop terminate naturally via stop_reason; keep iteration caps as a safety net only
✗ Prompt-based enforcement for critical business rules CRITICAL
✓ Use programmatic hooks (PreToolUse/PostToolUse) for deterministic enforcement
✗ Sentiment-based escalation to human agents HIGH
✓ Escalate on policy gaps, capability limits, explicit requests, or business thresholds
✗ Self-reported confidence scores for decision-making HIGH
✓ Use structured criteria and programmatic checks based on observable facts
Domain 2 · Tool Design & MCP
✗ Generic error messages ("Operation failed") CRITICAL
✓ Return structured errors: isError, errorCategory, isRetryable, and context
✗ Silently returning empty results for access failures CRITICAL
✓ Distinguish access failures (isError: true) from genuinely empty results
✗ Giving one agent 18+ tools HIGH
✓ Keep 4-5 tools per agent; distribute the rest across specialized subagents
✗ Hardcoding API keys in .mcp.json configuration CRITICAL
✓ Use ${ENV_VAR} environment variable expansion in MCP config
Domain 3 · Claude Code Configuration
✗ Putting personal preferences in project-level CLAUDE.md MEDIUM
✓ ~/.claude/CLAUDE.md for personal prefs; .claude/CLAUDE.md for team standards
✗ Using commands for complex tasks that need context isolation HIGH
✓ Use skills with context: fork and allowed-tools restrictions
✗ Same-session self-review in CI/CD pipelines CRITICAL
✓ Use separate sessions for code generation and code review
Domain 4 · Prompt Engineering
✗ Vague instructions like "be thorough" or "find all issues" CRITICAL
✓ Provide explicit, measurable criteria (e.g., "flag functions exceeding 50 lines")
✗ Assuming tool_use guarantees semantic correctness HIGH
✓ tool_use guarantees STRUCTURE only — validate values with business-rule checks
✗ Generic retry messages ("There were errors, please try again") HIGH
✓ Append specific error details: which field, what was wrong, expected vs actual
Domain 5 · Context & Reliability
✗ Progressive summarization of critical customer details CRITICAL
✓ Use immutable "case facts" blocks positioned at the start of context
✗ Aggregate accuracy metrics only (e.g., "95% overall") CRITICAL
✓ Track accuracy per document type (stratified metrics)
✗ No provenance tracking for multi-agent data HIGH
✓ Track source, confidence level, timestamp, and agent ID for all data
10 Critical 7 High 1 Medium 18 Total
§ 03 — Begin

Practice Test draws four of the six exam scenarios at random, mirroring the real exam structure. Official Questions runs through the canonical community practice set (sourced from claudecertifications.com), grouped by domain. Docs Deep Dives tests the same exam concepts at higher precision, synthesized from platform.claude.com documentation. Anti-Pattern Spotter drills the 18 cheatsheet anti-patterns — one question per pattern to build instant recognition. Intro to Agent Skills covers the first of four Anthropic-suggested courses: SKILL.md contract, priority order, progressive disclosure, and Skills vs CLAUDE.md / Subagents / Hooks. Scroll up to review the Anti-Patterns Cheatsheet for a quick-reference of all 18 patterns before starting a test. Explanations appear immediately after each answer. The question pool persists for this browser session only — no data is stored.

Question 1 of 24 0 correct
Scaled Score
0
of 1000 possible
Pass
Correct
0/24
Accuracy
0%
Time
—

Per-Scenario Breakdown

Review — Missed Questions