Pick the 4-week plan if…
- You ship with the Claude API or Claude Code in your day job and need 2-3 hours/day on weekdays plus a longer weekend block
- You want to walk in confident about all 5 domains, not just the two you already know cold
- You can hit at least one full 60-question mock before exam day and have a week to close gaps
- You're targeting a scaled score above 800 🟠, not just the 720 floor 🟡
- You need structured day-by-day pacing for week 1 and theme-based weekly summaries after that
- Your exam date is roughly 4 weeks out, give or take a few days
Otherwise see study plans index for the 2-week sprint or 12-week comprehensive variants.
Foundations and diagnostic
Week 1 is about calibration. You take the diagnostic on day 1 to expose your real gaps, then read the foundational frameworks that every later domain references. By Friday you should be able to draw the agentic loop on a whiteboard, name the 4 stop_reason values 🟡, and explain why subagents never inherit context.
Day 1Diagnostic and domain orientation~3 hr
- practiceTake the 30-question diagnostic~60 min
Sit it cold, no notes, 60 minutes max. Note which of the 5 domains 🟡 you bombed. Don't review answers yet; just record raw scores per domain.
- readRead the exam guide overview~30 min
Internalize the format: 60 MCQs, 120 minutes, 720/1000 to pass, no guessing penalty 🟡. Skim the domain weights so you know D1 (27%) and D3 (20%) carry the most leverage 🟠.
- knowledgeRead the AI Fluency framework (4D)~45 min
The 4D framework (Delegation, Description, Discernment, Diligence) is the conceptual spine the exam uses to frame agentic decisions. You'll see it again in scenarios.
- conceptCross-reference the 4D concept page~30 min
Map the framework to your diagnostic mistakes. Most failed D1 questions trace back to a Delegation or Discernment error, not a tool-mechanics error.
Day 2Claude 101 and API foundations~3 hr
- knowledgeRead Claude 101~45 min
Anchors model selection, message structure, and the basic prompt-and-response loop. Even if you've shipped with Claude for a year, do this; the exam tests the canonical Anthropic framing, not your shop's framing.
- knowledgeRead Claude API foundations~60 min
Focus on
messages.create, system prompts, multi-turn structure, and wherestop_reasonlives in the response. This is the bedrock for every D1 question. - conceptDrill the 4 stop_reason values~30 min
`end_turn`, `tool_use`, `max_tokens`, `stop_sequence` 🟡. Memorize what each implies for control flow. The exam will hand you a transcript and ask which value fired.
- conceptReview system prompts~30 min
Understand the system prompt as the deterministic anchor. Note the 'persona vs. instructions vs. case-facts' separation that case-facts-block doctrine builds on.
Day 3Agentic loops and the orchestration spine~3 hr
- conceptMaster the agentic loop~45 min
The standard loop: send messages, check
stop_reason, iftool_usethen execute and appendtool_result, repeat untilend_turn. This is the single most-tested mechanic across D1 (27%) 🟠. - knowledgeRead the subagents intro~45 min
Understand the ==hub-and-spoke== orchestration pattern and the rule that subagents never inherit parent context 🟡. This single fact is worth at least 2 exam questions.
- conceptCross-reference the subagents concept~30 min
Note when to spawn a subagent (parallel research, isolated context) vs. when to keep the work in the main loop (sequential dependency, shared state).
- scenarioWalk the Multi-Agent Research scenario~60 min
Read the P3.3 🟡 walkthrough end to end. Don't take the questions yet; just absorb how the architecture is justified.
Day 4Context window and capabilities~2.5 hr
- knowledgeRead AI Capabilities and Limitations~45 min
Learn where Claude struggles: long arithmetic chains, exact recall from the middle of a 200K context, deterministic JSON without schema enforcement. The exam tests calibrated trust, not Claude evangelism.
- conceptDrill context window mechanics~30 min
Internalize the lost-in-the-middle effect 🟡 and the progressive-summarization trap. Both show up in D5 (15%) 🟠questions.
- conceptReview prompt caching~30 min
Know when
CacheControlEphemeralsaves cost vs. when it silently invalidates. Cache invalidation is a favorite trap question. - knowledgeRead Claude Cowork intro~30 min
Frames the human-in-the-loop pattern that D5 reliability questions reference. Pay attention to the structured handoff format.
Day 5Week 1 consolidation and mini-mock~2.5 hr
- practiceTake a 15-question D1 quiz~30 min
D1 carries 27% weight 🟠, which is more than any other domain. If you're below 70% here, plan extra D1 time in week 2.
- scenarioWalk the Customer Support scenario (P3.1)~45 min
The canonical 'first scenario' in the official guide 🟡. Notice how the architecture choices map to D1 + D5 reliability concerns.
- readSkim the anti-pattern reference~30 min
Just the headings for now. This becomes mandatory reading in week 4, but seeing the failure-mode catalog early reframes the rest of your prep.
- conceptReview escalation patterns~30 min
Closes the week on a D5 note. Know when to escalate to a human and what the structured handoff packet contains.
D1 Agentic Architecture and D2 Tool Design
Week 2 is the heaviest content week. D1 (27%) + D2 (18%) = 45% 🟠of the exam, so spend the time. By Friday you should be able to design a tool schema cold, name all 4 isError categories 🟡, and explain why the 18-tool degradation point matters for tool-set design.
D1 deep dive: orchestration and stop control
Cement the agentic loop, subagent rules, and the role of stop_reason as the loop's traffic signal. Practice tracing transcripts and predicting which stop_reason fires next.
D2 foundations: tool design as deterministic selector
Tool descriptions are the ==deterministic selector== that drives Claude's tool choice 🟡. Master schema clarity, the 4 isError categories, and the tool_choice modes (auto, any, specific tool, none).
D2 advanced: MCP integration
MCP shows up in roughly a quarter of D2 questions. Read both MCP knowledge pages back to back, then drill the ==18-tool degradation== rule 🟠and when to split a tool server.
- knowledgeMCP Foundations
- knowledgeMCP Advanced
- conceptMCP concept page
- practice30-question D1+D2 mock
D3 Claude Code and D4 Prompt Engineering
Another 38% of the exam 🟠sits in D3 + D4. Week 3 is where Claude Code stops being 'a tool you use' and starts being 'a system you can configure for someone else'. By Friday you should be able to write CLAUDE.md hierarchy from memory, design a Skill with proper YAML frontmatter, and explain the validation-retry loop for structured output.
D3 mechanics: Claude Code configuration
Master the ==CLAUDE.md hierarchy== (project > user > enterprise) 🟡, YAML frontmatter globs for file scoping, Plan Mode triggers, and the -p headless flag for CI/CD pipelines.
- knowledgeClaude Code 101
- knowledgeClaude Code in Action
- conceptCLAUDE.md hierarchy
- conceptPlan Mode
- conceptHooks: PreToolUse and PostToolUse
D3 advanced: Skills and workflows
Agent Skills are the highest-leverage D3 topic. Internalize Skills frontmatter, when to invoke a skill vs. inline the work, and the relationship between Skills and subagents.
- knowledgeAgent Skills intro
- conceptSkills concept page
- scenarioCode Generation with Claude Code (P3.2)
- scenarioClaude Code in CI/CD (P3.5)
- scenarioAgent Skills for Developer Tooling (P3.12)
D4: prompt engineering and structured output
The validation-retry loop is the canonical D4 pattern: validate the JSON, on failure send the validator's error back to Claude as a tool_result, retry. Few-shot examples drive determinism; ==attention engineering== drives recall under long context.
- conceptPrompt engineering techniques
- conceptAttention engineering
- conceptEvaluation and eval design
- conceptStructured outputs (revisit)
- practice30-question D3+D4 mock
D5 Reliability, scenario depth, and final mock
Week 4 is synthesis week. D5 is only 15% 🟠of the exam, but it shows up in nearly every scenario question, so you fold it in while walking all 13 scenarios. By Sunday you sit a full 60-question mock, close any gaps, and lock in your readiness call.
D5 reliability and context management
Master the ==case-facts-block== doctrine, structured session state, and checkpoint patterns. Understand why the progressive-summarization trap silently corrupts long sessions and what to do about it.
- conceptCase-facts-block doctrine
- conceptSession state management
- conceptCheckpoints
- knowledgeClaude with Bedrock (deployment angle)
- knowledgeClaude with Vertex (deployment angle)
All 13 scenarios in one pass
The official guide names 6 scenarios; the real exam draws from ==13== 🟢. Walk the remaining ones you haven't deep-read, including the 2 confirmed beyond-guide scenarios.
Full mock and gap closure
Mid-week: take a fresh 60-question mock under exam conditions (120 minutes, no notes). Spend the back half of the week reviewing every miss against the anti-pattern reference and re-reading the relevant concept page.
Which Knowledge mirrors to read each week
The Knowledge pillar mirrors all 13 Anthropic Skilljar courses. We map them across the 4 weeks so you don't over-read foundations or skip deployment material.
| Week | Knowledge pages |
|---|---|
| Week 1 | ai-fluency-framework · claude-101 · claude-api-foundations · ai-capabilities-limitations · claude-cowork-intro · subagents-intro |
| Week 2 | mcp-foundations · mcp-advanced |
| Week 3 | claude-code-101 · claude-code-in-action · agent-skills-intro |
| Week 4 | claude-with-bedrock · claude-with-vertex |
2-week sprint and 12-week comprehensive
The 4-week plan is canonical. The two alternates use the same content at different paces — they are not separate plans.
2-week sprint
You already operate Claude in production across agentic loops, MCP, and Claude Code, and your exam is locked under 14 days out. You need the compressed path that skips redundant fundamentals.
Collapse week 1 into 2-3 days of diagnostic plus targeted reading on weak domains only. Spend the next 5-6 days on D1 + D2 + D3 deep dives plus all 13 scenarios. Reserve the final 3-4 days for one full mock, gap closure, and a second 30-question mock. Skip foundational knowledge pages like AI Fluency and AI Capabilities & Limitations if your diagnostic shows mastery.
12-week comprehensive
You're newer to Claude or coming from a different LLM stack, you want certification plus durable mastery, or you're scheduling around heavy work commitments at 4-6 hours/week.
Stretch each domain into 2 weeks of reading plus building. Add 2 weeks for original projects (a working subagent system, a custom MCP server, a real Claude Code workflow with hooks). Use weeks 11 and 12 for two full mocks and scenario walkthroughs. The 12-week plan adds a final 'shadow build' phase where you implement one full reference architecture end to end.
Before you book the proctored exam
Don't book the seat until you can clear this gate cold. The exam fee resets and re-takes burn momentum, so spend the extra 2-3 days if anything below is shaky.
- Hit ≥720/1000 🟡 on a fresh 60-question mock taken under timed conditions, with no notes and no second attempt
- Score ≥70% on each individual domain in your most recent mock (no domain dragging the average up)
- Walked all ==13 scenarios== 🟢, including the 2 beyond-guide ones (P3.10 Conversational AI Patterns, P3.11 Agent Skills for Enterprise KM)
- Reviewed every entry in the /reference anti-pattern catalog at least once and can name the 5 highest-leverage failure modes
- Can recite from memory: the 4
stop_reasonvalues, the 4isErrorcategories, the 4tool_choicemodes, and the 5 domain weights - Can draw the agentic loop on a whiteboard and annotate where each
stop_reasonshort-circuits it - Have a written 'first 10 minutes' plan for the exam: which domain to triage first, when to flag, when to skip
- Logistics confirmed: proctor software installed, ID ready, quiet room booked, 2.5-hour block held with a 30-minute buffer
Once your Skilljar entitlement lands
Once you've booked the seat, the prep shape changes. The next two weeks are about ==consolidation, not expansion==: tighten what you know, don't open new fronts.
Lock the calendar
Block the exam window plus 30 minutes either side. Block 2 hours the morning of for a final scenario re-read and a light D1 warmup. Tell your team you're unavailable; protect the slot.
Switch from learning to drilling
Stop opening new knowledge pages. Re-read your own notes and the ==/reference anti-pattern catalog==. Take one 30-question mock every 3 days, focused on whichever domain trailed in your last full mock.
Build a personal cheat-sheet (for studying, not the exam)
One page: 5 domain weights, 4 stop_reason values 🟡, 4 isError categories, 4 tool_choice modes, CLAUDE.md hierarchy, the 13 scenario names. Read it daily until you can rewrite it from memory in under 5 minutes.
Plan the exam-day mechanics
Decide your triage rule. Most candidates do best by skipping any question that needs more than 90 seconds on first pass, then returning. ==No guessing penalty== 🟡 means you answer everything before time expires; never leave a blank.
Plan the post-exam move
Results take ==2 business days per the official guide, often 7-10 in practice== 🟠. Decide now what you'll do with the credential: update LinkedIn the day results land, draft a short post about what you'd tell yourself 4 weeks ago, file the badge artifact in your portfolio.