CCA-F Prep — Training Plan v3

Claude Certified
Architect — Foundations

7-day intensive training plan targeting the CCA-F exam — Anthropic's first official technical credential, launched March 12, 2026. All 5 exam domains covered: agentic architecture (27%), tool design & MCP (18%), Claude Code config (20%), prompt engineering (20%), context & reliability (15%). Built for Fynaptix UAE consulting positioning.

7Training Days
60Exam Questions
5Exam Domains
CCA-FEnd Goal
Five CCA-F exam domains — mapped to each day
🤖
27%
Agentic Architecture
Multi-agent systems, hub-and-spoke, task decomposition, orchestration loops
🔧
18%
Tool Design & MCP
MCP server/client design, tool schemas, transport mechanisms, tool boundaries
⌨️
20%
Claude Code Config
CLAUDE.md hierarchies, slash commands, CI/CD integration, hooks
✍️
20%
Prompt Engineering
JSON schemas, few-shot techniques, structured outputs, validation retry loops
🧠
15%
Context & Reliability
Long-context management, handoff patterns, confidence calibration, RAG
📍 Your Progress
0 of 7 days complete — click a circle to mark done
Daily schedule
01
Agentic Architecture & Orchestration
Exam Domain 1 (27%) · Agentic loops · hub-and-spoke · task decomposition · failure modes
~3.5 hrs
+
Morning
75 min
Agentic Loops & Orchestration Theory
Domain 1 is the largest exam domain at 27%. It focuses on how you design systems where Claude acts autonomously over multiple steps. The core loop: perceive → plan → act → observe → repeat. The exam tests architectural judgment, not just feature knowledge.
📄 Claude Agents Documentation 📄 Tool Use & Agentic Patterns 🎓 Introduction to Subagents (Academy) 📝 CCA-F Free Study Guide
🎯 Exam Focus — Domain 1 (27%)
Expect scenario questions like: "A customer service agent produces inconsistent responses after 8 turns — which architectural change most likely resolves this?" Answer: context dilution + summarisation handoff pattern. Architecture judgment beats feature recall.
  • Learn the 4 agentic failure modes: context dilution, hallucinated tool calls, infinite loops, permission escalation
  • Understand hub-and-spoke: one orchestrator Claude spawning specialist sub-agents for parallel workstreams
  • Know when to use single-agent vs multi-agent: based on complexity, parallelisation, and isolation needs
  • Understand the roles: orchestrator Claude vs worker Claude instances — different prompts, different tool access
Midday
75 min
Build a Hub-and-Spoke Multi-Agent System (Python)
Hands-on building is the fastest way to internalise this domain. Build a research system: orchestrator Claude receives a query → decomposes into 3 subtasks → dispatches 3 parallel worker Claudes via API → synthesises results. This exact pattern is exam-tested.
⭐ Anthropic Cookbook (agent examples) 📄 Claude API Quickstart
  • Orchestrator: receives topic → generates 3 JSON subtasks → dispatches to workers via concurrent API calls
  • Workers: each gets one subtask + web search tool → returns structured JSON result
  • Orchestrator: receives 3 results → synthesises into final answer with per-source attribution
  • Test failure: what happens when one worker errors? Build graceful degradation logic.
Afternoon
60 min
Exam-Style Scenario Practice
The exam uses 2-3 paragraph scenarios before each question. Practice the thinking pattern: given a system description + a symptom, identify the root cause and the architectural fix.
📝 CertSafari — 611 Practice Questions
  • Practice: given "analyse Q3 sales and draft an exec summary", decompose into 4 agentic steps
  • Practice: design a code review agent that flags bugs, checks style, generates a PR comment
  • Memorise: sub-agent instances without prior generation context are more effective reviewers than self-review
  • Do 10 Domain 1 practice questions from CertSafari — review every wrong answer in detail
📦 Day 1 Deliverable
Working hub-and-spoke multi-agent Python script committed to GitHub. Handwritten cheat sheet: 4 failure modes, orchestrator/worker pattern, task decomposition steps, single vs multi-agent decision criteria. Revise on exam day.
02
Tool Design & MCP Integration
Exam Domain 2 (18%) · MCP server/client · tool schemas · transport mechanisms · tool boundary design
~3.5 hrs
+
Morning
75 min
MCP Architecture Deep Dive
The CCA-F expects you to design MCP servers, not just use them. Understand the full client/server/transport model, how tools are registered, and why tool schema design directly impacts reliability. This is infrastructure-level thinking.
📄 MCP Official Documentation 🎓 Intro to MCP (Academy) 🎓 MCP Advanced Topics (Academy) ⭐ MCP Reference Servers
🎯 Exam Focus — Domain 2 (18%)
Key exam concept: tool boundary design. Too many tools overwhelm Claude's reasoning (tool selection overhead). Too few tools and they're too broad to invoke correctly. The exam tests knowing when to split or merge tools, how tool descriptions drive correct invocation, and how to handle tool errors without stalling the agent.
  • Know the 3 MCP transport mechanisms: stdio (local), SSE (remote persistent), streamable HTTP (remote stateless)
  • Understand tool input schema design: required vs optional fields, enum constraints, description quality
  • Know MCP primitive types: tool (actions), resource (data), prompt (template) — when to use each
  • Understand sampling: MCP servers can request completions from the host Claude — know why this matters
Midday
90 min
Build a Production MCP Server (TypeScript)
Build a Fynaptix-branded MCP server wrapping a business API — a CRM or calendar integration with 3-5 well-designed tools. This is also a portfolio piece for UAE client demos.
⭐ MCP TypeScript SDK 📄 Claude MCP Integration Guide
  • Server: expose get_contact, create_task, log_interaction — 3 distinct, non-overlapping tools
  • Each tool: tight JSON schema with required fields, clear description, structured error response
  • Register locally with Claude Desktop — test each tool from a live conversation
  • Document your tool boundary decisions: why 3 tools, not 1 broad tool or 10 micro-tools
Afternoon
45 min
Tool Design Anti-Patterns & Practice Questions
Tool design quality is tested through architectural scenarios. Internalise the anti-patterns — the exam will present them as "what's wrong with this design" questions.
  • Anti-pattern: one do_everything tool with 20 parameters → Claude cannot invoke reliably
  • Anti-pattern: 30 micro-tools → excessive selection overhead every turn
  • Sweet spot: 5–12 tools, clear non-overlapping responsibilities, vivid descriptions
  • Do 10 Domain 2 practice questions — focus on MCP transport type and schema design questions
📦 Day 2 Deliverable
Working TypeScript MCP server with 3+ tools and JSON schemas committed to GitHub. README explaining architectural decisions. Cheat sheet: 3 MCP transport types, 3 primitive types, sampling pattern, tool boundary rules.
03
Claude Code Configuration
Exam Domain 3 (20%) · CLAUDE.md hierarchies · slash commands · hooks · CI/CD integration
~3.5 hrs
+
Morning
75 min
CLAUDE.md — The Persistent Tech Lead
CLAUDE.md is one of the most tested Claude Code concepts. The exam frames it as the "technical lead in a file" — encoding architectural rules, naming conventions, coding standards, and business context that Claude follows throughout a project. Master the full hierarchy.
📄 Claude Code Memory & CLAUDE.md 🎓 Claude Code 101 (Academy) 🎓 Claude Code in Action (Academy)
🎯 Exam Focus — Domain 3 (20%)
The exam tests CLAUDE.md hierarchy: ./CLAUDE.md overrides parent directory which overrides ~/.claude/CLAUDE.md. Also tested: slash command file location (.claude/commands/), hook pre/post execution behaviour, and running Claude Code SDK in non-interactive mode for CI/CD pipelines.
  • Memorise hierarchy: ./CLAUDE.md → parent dirs → ~/.claude/CLAUDE.md → @include imports
  • Know what belongs in CLAUDE.md: architecture rules, naming conventions, tool preferences, forbidden patterns
  • Understand @include path/to/file.md syntax for shared context across projects
  • Write 3 CLAUDE.md files: Fynaptix n8n project, UAE fintech client, Python data pipeline
Midday
75 min
Slash Commands & Hooks
Slash commands and hooks are power features that compound Claude Code's value in enterprise workflows. Commands encode complex multi-step operations. Hooks enforce governance without relying on conversation instructions.
📄 Slash Commands Reference 📄 Claude Code Hooks
  • Create /deploy:staging command: runs tests → builds → deploys → rollback on failure
  • Create a pre-tool hook: blocks file writes to /production without explicit confirmation token
  • Create a post-tool hook: logs every tool call + result to a structured JSON audit file
  • Understand hook failure modes: hook error → tool call blocked (pre) vs warning logged (post)
Afternoon
60 min
CI/CD Integration & Non-Interactive Mode
The exam tests Claude Code in automated pipelines — running without human input. This is the Claude Code SDK in non-interactive mode, used for automated code review, test generation, or documentation on every PR.
📄 Claude Code SDK 📄 GitHub Actions Integration
  • Build a GitHub Action: on PR open → Claude Code reviews diff → posts structured review comment
  • Use claude --non-interactive --output-format json — the CI/CD pattern for machine-readable output
  • Understand: non-interactive mode + JSON output format = the exam's CI/CD answer
  • Do 10 Domain 3 practice questions — focus on CLAUDE.md hierarchy and hooks questions
📦 Day 3 Deliverable
3 CLAUDE.md files (Fynaptix project, UAE fintech, Python pipeline). Working slash command + pre/post hook pair. GitHub Action for automated PR review using non-interactive mode. CLAUDE.md hierarchy diagram for exam revision.
04
Prompt Engineering & Structured Output
Exam Domain 4 (20%) · JSON schemas · validation loops · few-shot · prefill · extended thinking
~3.5 hrs
+
Morning
75 min
Structured Output Engineering
The exam tests production-grade prompt engineering — reliably extracting structured data from Claude in agentic pipelines. This is not creative prompting; it's reliability engineering. JSON schema enforcement, retry loops, and output validation are the core skills.
📄 Prompt Engineering Overview 📄 Structured Outputs 🎓 Prompt Engineering Fundamentals (Academy)
🎯 Exam Focus — Domain 4 (20%)
Exam scenario: an agentic pipeline fails intermittently because Claude sometimes returns markdown-wrapped JSON. Fix: explicit JSON schema in system prompt + prefill the assistant turn with { + a validation-retry loop that sends the parse error back as a user message and retries up to 3 times.
  • Master the prefill technique: start the assistant's turn with {" to force JSON output without preamble
  • Build a validation-retry loop: parse output → on failure, inject error as user message → retry up to 3×
  • Understand few-shot prompting: 2-3 input/output examples in the system prompt dramatically improve output consistency
  • Know when to use extended thinking: complex multi-step reasoning — NOT structured extraction (latency cost)
Midday
75 min
Build a Reliable Document Extraction Pipeline
Build a pipeline extracting structured data from unstructured documents (invoices, contracts, emails) with 99%+ reliability. This is a real Fynaptix use case and directly maps to CCA-F exam scenarios.
⭐ Anthropic Cookbook (extraction examples)
  • System prompt: JSON schema definition + 2 few-shot examples of correct extraction
  • Validation loop: parse → validate against schema → on failure, inject specific error + retry
  • After 3 failures: flag for human review — never silently fail in production
  • Test with 5 deliberately malformed inputs — measure retry rate and final success rate
Afternoon
60 min
Advanced Techniques & Exam Practice
Master the remaining Domain 4 techniques: XML tags for context separation, chain-of-thought elicitation, and self-consistency voting.
  • XML tags: use <context>, <instructions>, <examples> to separate prompt sections — improves attention
  • Chain-of-thought: "Think step by step before answering" → measurably improves reasoning quality
  • Self-consistency: generate 3 independent answers → take majority vote → reduces factual hallucination
  • Do 10 Domain 4 practice questions from CertSafari
📦 Day 4 Deliverable
Working Python extraction pipeline with validation-retry loop (GitHub-committed). Reusable system prompt template for structured extraction. Domain 4 reference card: prefill technique, retry pattern, few-shot structure, when to use/avoid extended thinking.
05
Context Management & Reliability
Exam Domain 5 (15%) · Long-context · summarisation handoff · RAG · confidence calibration
~3 hrs
+
Morning
75 min
Long-Context Architecture & Handoff Patterns
Context window management is critical in agentic systems that run for many turns. The exam tests context dilution recognition and the architectural patterns to fix it — primarily summarisation handoffs.
📄 Context Window Management 📄 Long Context Best Practices 🎓 Building with Claude API (Academy)
🎯 Exam Focus — Domain 5 (15%)
Exam pattern: a long-running agent produces lower-quality responses after many turns. Root cause is context dilution. Solution: implement a summarisation handoff — Claude summarises state → new instance starts fresh with the summary as context. Also know "lost in the middle": information at the centre of very long contexts receives less attention.
  • Know context window limits: Claude Sonnet 4.6 = 200K tokens in, 8K out; Claude Opus 4.6 = 200K in, 32K out
  • Understand attention dilution — information in the middle of long contexts is attended to less reliably
  • Build a summarisation handoff: detect context usage >80% → trigger summarisation → restart with summary as context
  • Know the "lost in the middle" finding — place critical instructions at the start OR end of long prompts
Midday
60 min
Self-Correction & Confidence Calibration
Production agentic systems need mechanisms for self-correction without human intervention. Multi-instance review and confidence scoring are the exam-tested patterns for this.
  • Multi-instance review: separate Claude for generation and review — reviewer has no generation context (no reasoning bias)
  • Per-file review passes: for large codebases, review file-by-file then do a cross-file integration pass
  • Confidence scoring: prompt Claude to output {"confidence": 0.87, "answer": "..."} → route low-confidence to human review
  • Never self-review in the same session: the model retains its own reasoning context and is less likely to catch its own errors
Afternoon
45 min
RAG Architecture & Practice Questions
Retrieval-Augmented Generation (RAG) is tested as a context management decision — when to retrieve vs when to use long-context directly.
  • RAG vs long-context decision: RAG for large knowledge bases; long-context for document analysis tasks
  • Know chunking strategies: semantic chunking vs fixed-size, and why chunk overlap matters for recall
  • Understand re-ranking: retrieve 20 chunks → re-rank to top 5 → send to Claude (reduces noise)
  • Do 10 Domain 5 practice questions from CertSafari
📦 Day 5 Deliverable
Summarisation handoff script (detects context usage → summarises → restarts). Multi-instance review pattern implementation. Domain 5 cheat sheet: context limits, handoff pattern, confidence calibration thresholds, RAG vs long-context decision tree.
06
Full Integration — Production System Design
All 5 domains · end-to-end enterprise scenario · Fynaptix UAE portfolio capstone
~4 hrs
+
Morning
90 min
Design a Production System — Fynaptix UAE Onboarding Agent
Design a complete production system touching all 5 exam domains. Scenario: Fynaptix is building an AI-powered client onboarding system for a UAE fintech. System intake documents, extracts structured data, verifies against regulatory databases, drafts an onboarding report, and routes edge cases for human review.
  • Domain 1: Hub-and-spoke design — orchestrator + 3 specialist agents (document extraction, regulatory check, report drafting)
  • Domain 2: MCP server with tools: parse_document, check_kyc_database, draft_report, flag_human_review
  • Domain 3: CLAUDE.md encoding UAE KYC requirements, required output fields, permitted data sources
  • Domain 4: Structured extraction prompt with JSON schema + validation retry loop for document data
  • Domain 5: Confidence scoring on regulatory lookups — confidence <0.85 → human review queue
Midday
90 min
Build the Core Pipeline
Implement the core of the UAE onboarding agent. Even a simplified demonstration of the architecture is a compelling portfolio piece for Fynaptix UAE enterprise consulting.
⭐ Anthropic Cookbook (production examples)
  • Build orchestrator + document extraction agent: takes PDF text → returns structured KYC JSON
  • Build MCP server with the 4 tools above, JSON schemas, structured error responses
  • Write CLAUDE.md for the project: UAE regulatory context, confidence thresholds, output format requirements
  • Test with 3 sample onboarding documents — measure extraction accuracy and confidence score distribution
Afternoon
60 min
Architecture Diagram & Write-Up
Document the system architecture. The exam presents architecture diagrams and asks you to identify issues or improvements — creating them builds the spatial intuition you need.
  • Draw full architecture: data flow, agent roles, MCP tool calls, confidence routing, human-in-the-loop
  • Write 1-page architectural rationale: design choices and their tradeoffs
  • Identify 3 failure modes in your design and their mitigations
  • Draft LinkedIn post: "How I built a UAE fintech onboarding agent with Claude" — publish after passing the exam
📦 Day 6 Deliverable — Primary Portfolio Piece
Complete Fynaptix UAE Fintech Onboarding Agent: GitHub repo with orchestrator + MCP server + CLAUDE.md + architecture diagram + 1-page write-up. Primary CCA-F credential demo and a Fynaptix UAE business development asset.
07
Full Mock Exam & Exam Scheduling
60-question timed simulation · weak domain remediation · Partner Network · book the exam
~4 hrs
+
Morning
120 min
Full 60-Question Timed Mock Exam
Simulate the actual exam: 60 questions, 120-minute timer, no notes, scenario-based MCQs. Use CertSafari's 611 practice questions. Aim for accuracy first, then speed. The real exam has 2-3 paragraph scenarios per question.
📝 CertSafari — 611 Practice Questions 📝 claudecertifications.com — 25 Free Questions
  • Set a 120-minute timer — answer all 60 questions without checking answers mid-exam
  • Mark uncertain questions (still answer — guess and flag for review)
  • After: score by domain — identify your weakest domain (most wrong answers)
  • Target: 80%+ overall, no domain below 75%
Midday
60 min
Weak Domain Remediation
Based on your mock exam score, spend focused time on your lowest-scoring domain. Review docs, re-read your cheat sheet, do 20 more domain-specific practice questions.
  • Domain 1 weak: re-read agentic docs + review Day 1 cheat sheet + re-trace your multi-agent code
  • Domain 2 weak: re-read MCP transport docs + review your server's tool schema decisions
  • Domain 3 weak: re-read CLAUDE.md hierarchy docs + re-test your slash commands and hooks
  • Domain 4 weak: rebuild your extraction pipeline with a harder malformed-input test case
  • Domain 5 weak: re-read context window docs + trace through your summarisation handoff script
Afternoon
60 min
Partner Network Application & Exam Booking
The CCA-F requires Claude Partner Network membership. Fynaptix qualifies immediately as an organisation bringing Claude to market — any such org is eligible. Apply now (free), receive approval in ~1-2 weeks, then schedule on Skilljar.
🏢 Claude Partner Network — Apply Free 📝 CCA-F Exam Access Request 🎓 Anthropic Academy (Skilljar)
  • Apply to Partner Network now via claude.com/partners — free, Fynaptix qualifies immediately
  • While awaiting approval: complete the free Academy courses (no Partner access required)
  • On approval: request exam access → schedule on Skilljar ($99, or free in first 5,000 partner employees)
  • Prepare LinkedIn post and Fynaptix.com update for credential — publish immediately on passing
📦 Week Complete
Mock exam score 80%+. Partner Network application submitted. Exam scheduled. GitHub portfolio with 4 projects (multi-agent system, MCP server, GitHub Action CI/CD, UAE onboarding agent). 5 domain cheat sheets ready for exam-day review. LinkedIn post drafted and ready to publish.
🎓 Anthropic Academy — CCA-F Exam Sequence
Every course is free, self-paced, and awards an Anthropic certificate on completion. Hosted at anthropic.skilljar.com. The sequence below is ordered by exam domain weight and dependency — start with Building with the Claude API.
~20hPriority Track
FreeAll Courses
5Domains Covered
OfficialCertificates
📋 Recommended Academy Sequence for CCA-F
1
Building with the Claude API
The flagship 8-hour course: messages, tool use, streaming, structured outputs, multi-turn conversations. Directly covers Domains 1, 2, and 4. Highest CCA-F return per hour — start here.
~8 hrs · Start immediately
→ Enrol free
2
Introduction to Subagents
Domain 1 (27%) — covers sub-agent patterns and multi-agent orchestration. Directly teaches hub-and-spoke architecture and parallel agent execution tested in exam scenarios.
~1.5 hrs · Week 1
→ Enrol free
3
Introduction to Model Context Protocol
Domain 2 (18%) — client/server architecture, transport mechanisms, tool primitives, MCP integration with Claude. Essential. Do not skip.
~3 hrs · Week 1
→ Enrol free
4
Claude Code 101
Domain 3 (20%) — CLAUDE.md configuration, slash commands, Claude Code architecture. CLAUDE.md hierarchy is among the most frequently tested Domain 3 topics.
~1.5 hrs · Week 2
→ Enrol free
5
Claude Code in Action
Domain 3 depth — CLAUDE.md hierarchies in detail, hooks, CI/CD integration, non-interactive mode SDK. The depth layer after Claude Code 101.
~1 hr · Week 2
→ Enrol free
6
Prompt Engineering Fundamentals
Domain 4 (20%) — structured outputs, XML tags, JSON schema enforcement, validation-retry loops. Covers the core exam-tested prompting patterns.
~2 hrs · Week 2
→ Enrol free
7
MCP: Advanced Topics
Domain 2 depth — sampling, notifications, file system access, advanced transport patterns. Required for building production MCP servers and troubleshooting complex integrations.
~3 hrs · Week 2
→ Enrol free
8
Claude 101
Broad Claude foundation — projects, artifacts, context management. Domain 5 support plus useful for Fynaptix client training sessions.
~2 hrs · Week 3
→ Enrol free
Full course catalogue — CCA-F relevance rated
🔴 Must Complete — Direct CCA-F exam coverage
Do these first
MUST
ACADEMY-01
Building with the Claude API
Messages API, tool use, streaming, structured outputs, multi-turn conversations. Highest CCA-F ROI per hour.
~8 hrsEnrol →
Domains 1, 2, 4 — ~65% of exam weight
MUST
ACADEMY-02
Introduction to Subagents
Sub-agent patterns, hub-and-spoke orchestration, parallel execution. Domain 1 is 27% — essential.
~1.5 hrsEnrol →
Domain 1 — Agentic Architecture (27%)
MUST
ACADEMY-03
Introduction to MCP
MCP client/server architecture, transport mechanisms, tool primitives, Claude integration.
~3 hrsEnrol →
Domain 2 — Tool Design & MCP (18%)
MUST
ACADEMY-04
Claude Code 101
CLAUDE.md configuration, custom slash commands, Claude Code architecture and workflows.
~1.5 hrsEnrol →
Domain 3 — Claude Code Config (20%)
MUST
ACADEMY-05
Prompt Engineering Fundamentals
Structured outputs, XML tags, JSON schema enforcement, validation retry loops, few-shot examples.
~2 hrsEnrol →
Domain 4 — Prompt Engineering (20%)
🟡 Should Complete — Depth layer
Week 2
SHOULD
ACADEMY-06
Claude Code in Action
CLAUDE.md hierarchies in depth, hooks, CI/CD integration, Claude Code SDK non-interactive mode.
~1 hrEnrol →
Domain 3 depth — hooks and CI/CD patterns
SHOULD
ACADEMY-07
MCP: Advanced Topics
Advanced MCP: sampling, notifications, file system access, advanced transport mechanisms.
~3 hrsEnrol →
Domain 2 depth — production MCP patterns
SHOULD
ACADEMY-08
Claude with Amazon Bedrock
Enterprise Claude on Bedrock — critical for UAE clients requiring data residency. Gulf regulated markets default to Bedrock.
~2 hrsEnrol →
Fynaptix UAE positioning — regulated deployment
⚪ Nice to Have — Post-exam context
Week 3+
OPTIONAL
ACADEMY-09
Introduction to Claude Cowork
Task loop, plugins, skills, file workflows. Enterprise product knowledge — not CCA-F exam content. Do after the exam if targeting Cowork configurator roles.
~2 hrsEnrol →
Post-exam — enterprise product knowledge
OPTIONAL
ACADEMY-10
AI Fluency: Framework & Foundations
The 4D Framework (Delegation, Description, Discernment, Diligence). Useful for Fynaptix client training sessions — not CCA-F exam content.
~2 hrsEnrol →
Fynaptix consulting delivery, not exam-critical
Claude Certified Architect — Foundations (CCA-F)
Anthropic's first official technical certification, launched March 12, 2026. A proctored 301-level exam validating ability to design and ship production-grade Claude AI systems at enterprise scale. This is a systems design exam — not a prompting or product knowledge test. Backed by Anthropic's $100M Partner Network investment.
Questions
60 MCQ
Duration
120 min
Format
Proctored
Level
301 (Architect)
Cost
$99 / Free*
Platform
Skilljar
Five exam domains — weight and key content
27%
Agentic Architecture & Orchestration
Multi-agent systems, hub-and-spoke design, task decomposition, failure modes, parallelisation, orchestrator/worker patterns
18%
Tool Design & MCP Integration
MCP server/client design, transport mechanisms (stdio/SSE/HTTP), tool boundary design, schema quality, sampling pattern
20%
Claude Code Configuration
CLAUDE.md hierarchies, @include syntax, slash commands, pre/post hooks, non-interactive mode, GitHub Actions integration
20%
Prompt Engineering & Structured Output
JSON schemas, prefill technique, few-shot examples, XML tags, validation-retry loops, extended thinking usage decisions
15%
Context Management & Reliability
Context window limits, attention dilution, summarisation handoffs, multi-instance review, confidence calibration, RAG patterns
Step-by-step certification path
1
Apply to Claude Partner Network (free, ~1-2 weeks)Fynaptix qualifies immediately — any organisation bringing Claude to market is eligible. Apply at claude.com/partners. Required for exam access.
2
Complete Academy Priority Track (~20 hours)8 priority courses in the sequence above. Free, no Partner access required. Directly maps to all 5 exam domains.
3
Build the 7-Day PortfolioComplete hands-on projects: multi-agent system, MCP server, Claude Code configs, extraction pipeline, UAE onboarding agent. Hands-on builds the architectural intuition the exam tests.
4
Score 80%+ on Mock Exam Before BookingUse CertSafari (611 questions) and claudecertifications.com (25 free). Score consistently above 80% across all domains before scheduling. Community benchmark: 985/1000 is achievable.
5
Request Exam Access & Schedule on SkilljarOnce Partner Network approved, request CCA-F exam access. Schedule on Skilljar. *First 5,000 Partner Network employees get it free; otherwise $99 per attempt.
6
Pass & Publish — LinkedIn + Fynaptix.comDigital badge for LinkedIn. Update Fynaptix.com homepage. Publish the prepared LinkedIn post. The CCA-F is becoming a baseline expectation at Accenture, Cognizant, Deloitte — an early-mover credential for Fynaptix in UAE.
Key resources
🏢 Apply: Partner Network 📝 CCA-F Exam Access 📝 CertSafari — 611 Questions 📝 claudecertifications.com (free guide + 25 Qs) 📄 LowCode Agency CCA-F Guide 📄 FlashGenius Domain Breakdown 📄 CertStud CCA-F Guide
Domain readiness — tick each before scheduling the exam
Domain 1 — 27%
Agentic Architecture
  • Can design a hub-and-spoke multi-agent system from scratch
  • Can name and explain all 4 agentic failure modes
  • Knows when to use single-agent vs multi-agent architecture
  • Can decompose a complex task into parallelisable subtasks
  • Working multi-agent Python system committed to GitHub
Domain 2 — 18%
Tool Design & MCP
  • Can explain all 3 MCP transport types and when to use each
  • Can design a well-bounded tool set (5–12 tools) for a system
  • Knows MCP primitive types: tool, resource, prompt
  • Understands the MCP sampling pattern
  • Working TypeScript MCP server committed to GitHub
Domain 3 — 20%
Claude Code Config
  • Can recite CLAUDE.md hierarchy from memory
  • Has written 3 CLAUDE.md files for different contexts
  • Can build and test a custom slash command
  • Understands pre/post hooks and their failure modes
  • Can run Claude Code in non-interactive mode for CI/CD
Domain 4 — 20%
Prompt Engineering
  • Can build a JSON schema + validation retry loop in Python
  • Knows the prefill technique and when to use it
  • Understands few-shot prompting for structured output
  • Knows when NOT to use extended thinking
  • Working extraction pipeline committed to GitHub
Domain 5 — 15%
Context & Reliability
  • Knows Claude Sonnet 4.6 context limits (200K in / 8K out)
  • Can implement a summarisation handoff pattern
  • Understands multi-instance review (reviewer has no gen context)
  • Can implement confidence scoring + human-review routing
  • Understands when to use RAG vs long-context
Academy
Course Progress
  • Building with the Claude API — completed + certificate
  • Introduction to Subagents — completed + certificate
  • Introduction to MCP — completed + certificate
  • Claude Code 101 — completed + certificate
  • Prompt Engineering Fundamentals — completed + certificate
  • Claude Code in Action — completed + certificate
  • MCP: Advanced Topics — completed + certificate
Portfolio
GitHub Projects
  • Hub-and-spoke multi-agent system (Python) — published
  • TypeScript MCP server with 3+ tools — published
  • 3 CLAUDE.md files with architecture rationale — published
  • Validation-retry extraction pipeline — published
  • UAE Fintech Onboarding Agent (Day 6 capstone) — published
Certification
CCA-F Roadmap
  • Applied to Claude Partner Network (Fynaptix)
  • Partner Network application approved
  • All 7 priority Academy courses complete
  • Mock exam score 80%+ across all domains
  • CCA-F exam scheduled on Skilljar
  • CCA-F exam passed 🏆
  • Credential live on LinkedIn + Fynaptix.com