The Claude Architect's Reference
Cert-Aligned — Claude Certified Architect: Foundations (D1–D5)
This volume maps 1:1 to Anthropic's Claude Certified Architect — Foundations five-domain taxonomy. It is exam-prep altitude: authoritative reference prose wrapped in light exam apparatus (objectives, watch-outs, self-checks), pointing outward to the Agentic Systems Design book where a domain needs design depth.
Start with the assessment test — a scored, cross-domain diagnostic that tells you which Parts to read closely and which to skim.
Domains
Part 1 · D1 — Agentic Architecture & Orchestration (27%)
- D1.1 Agentic Loops: stop_reason and Tool-Result Handling
- D1.2 Coordinator–Subagent Patterns: Hub-and-Spoke and Isolated Context
- D1.3 Subagent Invocation: AgentDefinition, the Agent Tool, and allowedTools
- D1.4 Multi-Step Workflows: Programmatic vs Prompt-Based Handoff
- D1.5 Agent SDK Hooks: Intercepting, Gating, and Normalizing the Loop
- D1.6 Task Decomposition: Sequential Pipelines vs Adaptive
- D1.7 Session State: resume, fork, and Scratchpads
Part 2 · D2 — Tool Design & MCP Integration (18%)
- D2.1 Effective Tool Interfaces: Descriptions, Boundaries, and Naming
- D2.2 Structured Error Responses: isError, Retryability, and the Protocol-Error Split
- D2.3 Tool Distribution and tool_choice: auto, any, Forced, and none
- D2.4 MCP Server Configuration: .mcp.json, Scopes, and Env-Var Expansion
- D2.5 Built-in Tools: The Roster, Execution Order, and Permission Gating
Part 3 · D3 — Claude Code Configuration & Workflows (20%)
- D3.1 CLAUDE.md Hierarchy & @import: Four Scopes That Concatenate
- D3.2 Slash Commands & Skills: Stored Prompts, Lazy-Loaded Capabilities
- D3.3 Path-Scoped Rules: Modular, Glob-Triggered Instructions
- D3.4 Plan Mode vs Direct Execution: Research Before You Edit
- D3.5 Iterative Refinement: The Loop, the Interview, and Test-Driven Prompting
- D3.6 CI/CD Integration: Headless Runs, Output Formats, and GitHub Actions
Part 4 · D4 — Prompt Engineering & Structured Output (20%)
- D4.1 Explicit Criteria over Vague Instructions
- D4.2 Few-Shot Prompting for Ambiguous Cases
- D4.3 Structured Output via Tool Use and JSON Schema
- D4.4 Validation, Retry, and Feedback Loops
- D4.5 Batch Processing: The Message Batches API
- D4.6 Multi-Pass Review: Independent Reviewers and Attention Dilution
Part 5 · D5 — Context Management & Reliability (15%)
- D5.1 Long-Conversation Context: Accumulation, Degradation, Compaction
- D5.2 Escalation and Ambiguity Resolution
- D5.3 Error Propagation Across Multi-Agent Systems
- D5.4 Large-Codebase Context: Compaction, Scratchpads, Delegation
- D5.5 Human Review and Confidence Calibration
- D5.6 Information Provenance: Citations and Temporal Validity
Exam objective coverage
Which chapters carry retrieval-practice questions in each domain — auto-derived from the question bank, so it can't drift from what's actually written. Work the full practice question bank (grouped by domain, answers behind a reveal).
| Domain | d1-01 | d1-02 | d1-03 | d1-04 | d1-05 | d1-06 | d1-07 | d2-01 | d2-02 | d2-03 | d2-04 | d2-05 | d3-01 | d3-02 | d3-03 | d3-04 | d3-05 | d3-06 | d4-01 | d4-02 | d4-03 | d4-04 | d4-05 | d4-06 | d5-01 | d5-02 | d5-03 | d5-04 | d5-05 | d5-06 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Agentic Architecture & Orchestration | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| Tool Design & MCP Integration | — | — | — | — | — | — | — | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| Claude Code Configuration & Workflows | — | — | — | — | — | — | — | — | — | — | — | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | — | — | — | — | — | — | — | — | — |
| Prompt Engineering & Structured Output | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | — | — | — |
| Context Management & Reliability | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
Status
All 30 chapters (D1.1–D5.6) are authored as self-contained study-guide chapters — each opens with a "Do I know this already?" diagnostic and closes with exam essentials. The question bank is seeded (2 per domain) and growing toward per-chapter coverage. See OUTLINE.md for the chapter plan and docs/BOOK-MAP.md for how this book relates to the others.