Claude Architect Foundations - O'Reilly Live Training

Contact: Website | LinkedIn | GitHub | O'Reilly | YouTube

Reference architectures, code examples, and practice scenarios for the Claude Architect role. This 4-hour O'Reilly Media live training is skills-first for Segments 1-3, then closes with a CCA-F certification capstone in Segment 4 (cert briefing + weighted practice questions). It teaches the production patterns that define a Claude Architect (agentic orchestration, tool design with MCP, Claude Code workflows, prompt engineering, context management) and gives you a runway to Anthropic's CCA-F exam.

Anthropic's Claude Certified Architect: Foundations (CCA-F) exam is currently restricted to Anthropic partners. The five reference files in this repo map to the published 5-domain exam blueprint, and CERT-PROGRAM-BRIEFING.md walks through exam mechanics, prep stack, and a week-before punchlist.

What is a Claude Architect?

The Claude Architect is an emerging job role focused on designing and building production-grade applications with Claude Code, the Claude Agent SDK, the Claude API, and Model Context Protocol (MCP). Organizations across the Claude Partner Network are hiring for this skillset.

Course at a glance

Segment	Duration	Topic	Key deliverable
1	50 min	Building AI Agents That Use Tools	Customer support agent with hook-enforced policy
2	50 min	Tool Design, Integration, and Claude Code Workflows	MCP config walkthrough + Claude Code hierarchy demo
3	50 min	Structured Output, Context, and Production Reliability	Invoice extractor with retry + triage scorecard
4	50 min	CCA-F Certification Capstone	Cert briefing + 10 weighted practice questions + take-home punchlist

Total: 4 hours (4 × 50-min segments + 3 × 10-min breaks). Instructors and learners should start at COURSE-FLOW.md.

CCA-F exam blueprint (the five core competencies)

Domain	Weight	Reference file	Focus
1 - Agentic Architecture & Orchestration	27%	domain-1-agentic.md	Agentic loops, multi-agent coordination, hooks, session management
2 - Tool Design & MCP Integration	18%	domain-2-tools-mcp.md	Tool descriptions, structured errors, scoped distribution, MCP config
3 - Claude Code Configuration & Workflows	20%	domain-3-claude-code.md	CLAUDE.md hierarchy, skills, slash commands, plan mode, CI/CD
4 - Prompt Engineering & Structured Output	20%	domain-4-prompts.md	Explicit criteria, few-shot prompting, JSON schemas via tool use, batch API
5 - Context Management & Reliability	15%	domain-5-context.md	Context preservation, escalation, error propagation, provenance

Exam mechanics: 60 multiple-choice questions, 120 minutes, scaled 100-1000 with 720 passing, proctored via ProctorFree, one attempt only, $99 (partner discount available). See CERT-PROGRAM-BRIEFING.md for the full briefing.

Repository layout

claude-architect/
├── INSTRUCTOR-SETUP.md         # Multi-day setup arc (machine config, env vars, repo clone, backup plans)
├── COURSE-FLOW.md              # Master instructor punchlist (4 segments × 50 min)
├── PRE-CLASS-CHECKLIST.md      # Instructor pre-flight (PowerShell)
├── CERT-PROGRAM-BRIEFING.md    # Segment 4 talk-track: exam mechanics, domain weights, week-before punchlist
├── PRACTICE-QUESTIONS.md       # 60-question community-sourced practice bank (cohort take-home)
├── practice-questions.json     # Machine-readable practice-question source
├── domain-1-agentic.md         # Reference: Agentic Architecture & Orchestration
├── domain-2-tools-mcp.md       # Reference: Tool Design & MCP Integration
├── domain-3-claude-code.md     # Reference: Claude Code Configuration & Workflows
├── domain-4-prompts.md         # Reference: Prompt Engineering & Structured Output
├── domain-5-context.md         # Reference: Context Management & Reliability
├── .mcp.json                   # Segment 2 Demo A anchor (4 servers, 3 transports)
├── hooks-example.py            # Agent SDK hooks: compliance enforcement
├── testing.md                  # Coordinator-subagent test patterns
├── scenario-cicd-integration.md # Codebase analysis skill with frontmatter
├── SKILL.md                    # Example slash-command / skill definition
├── CLAUDE.md                   # Claude Code project instructions for this repo
└── scripts/
    └── extract-practice-questions.py   # Build-time extractor for the practice-question files

Getting started

Prerequisites

Node.js 18+ for Anthropic SDK examples
Python 3.13+ for hooks and cookbook notebooks
Claude Code CLI installed and authenticated
An Anthropic API key set as ANTHROPIC_API_KEY

Setup

git clone https://github.com/timothywarner-org/claude-architect.git
cd claude-architect
npm install

Recommended learning path

Read COURSE-FLOW.md for the full 4-segment teaching arc.
Walk the five domain-*.md reference files in order. Each maps to a course segment and points at runnable cookbook notebooks.
Run the demo notebooks in your own environment. The course depends on Anthropic's official cookbooks (private/claude-cookbooks-main/).
Work through CERT-PROGRAM-BRIEFING.md and the PRACTICE-QUESTIONS.md bank if you're aiming at the CCA-F exam.
Build something. The reference architectures only land when you wire one of these patterns into a real workflow.

Practice scenarios (CCA-F exam pool)

Each scenario frames a realistic production context that a Claude Architect would encounter. The CCA-F exam draws 4 scenarios at random from a pool of 6:

#	Scenario	Primary competencies
1	Customer Support Resolution Agent	Domains 1, 2, 5
2	Code Generation with Claude Code	Domains 3, 5
3	Multi-Agent Research System	Domains 1, 2, 5
4	Developer Productivity Tooling	Domains 1, 2, 3
5	CI/CD Integration with Claude Code	Domains 3, 4
6	Structured Data Extraction Pipeline	Domains 4, 5

Key concepts quick reference

Agentic loop

Send request -> Check stop_reason -> "tool_use"? Execute tool, append tool_result, loop
                                  -> "end_turn"? Done
                                  -> "pause_turn"? Resume in next request

Tool selection

Descriptions drive selection. Detailed descriptions beat clever tool names.
Cap at 4-5 tools per agent. More tools degrade selection accuracy.
Structured errors as tool_result content. Never raise exceptions from tool implementations.

Claude Code configuration

User-level ~/.claude/CLAUDE.md for personal defaults
Project-level ./CLAUDE.md at repo root for team conventions
Subtree <subdir>/CLAUDE.md loads on demand
claude -p for headless / CI/CD usage

Structured output

tool_use with JSON schema = guaranteed schema compliance. Define output as a tool's input_schema, force the model to call it with tool_choice: {"type": "tool", "name": "..."}.
Few-shot examples > temperature. Two or three input-output pairs beat tuning sampling parameters.

Context and reliability

Pin case-facts at the top of long sessions
Summarize resolved turns; keep verbatim history only for the active issue
Escalate on explicit human request, policy gaps, or low confidence - never on sentiment alone

About the instructor

Tim Warner is a Microsoft MVP (Azure AI), Pluralsight Principal Author (200+ courses, 1M+ learners), Microsoft Press / Pearson senior content developer, and O'Reilly Live Learning instructor with 28+ years on the Microsoft stack. He teaches Feynman-style: first principles, no fluff, real demos.

Website: TechTrainerTim.com
LinkedIn: @timothywarner
GitHub: @timothywarner-org
YouTube: TechTrainerTim
O'Reilly: Author page

Disclaimer

This is an unofficial study guide built around Anthropic's publicly documented CCA-F exam blueprint. Practice questions in PRACTICE-QUESTIONS.md are community-sourced from Paul Larionov's study repo and are intended for calibration only, not as exam predictors. The authoritative source for exam content, registration, and policy is Anthropic (see the public CCA-F page and the Exam Policy). Use Anthropic's own Practice Exam to gauge readiness before scheduling.

Contributing

Contributions are welcome. See CONTRIBUTING.md for guidelines.

License

MIT - see LICENSE.

Found this useful? Open an issue with questions or feedback - I read every one.