Study Plan
Remember
- Mocks are easier than the real exam. Do not trust a high mock score.
- 5–10% of vendor mock questions are simply wrong. Verify against official docs only.
- Understand concepts. Do not memorize dumps — one changed word flips the answer.
Order of study
flowchart TD
A["1. This pack:<br/>exam/ + scenarios/ + patterns/"] --> B["2. Official docs<br/>(highest yield pages below)"]
B --> C["3. Question bank<br/>questions/ folder — understand every answer"]
C --> D["4. Mocks in exam mode<br/>with 120-min timer"]Highest-yield official doc pages
| Page | Why |
|---|---|
| Claude Code: headless / programmatic usage | Most flag questions come from here |
| Claude Code: CLI reference | Flags and defaults |
| Claude Code: permissions / IAM | Permission rules and modes |
| Claude Code: hooks | Hook events, guaranteed enforcement |
| Claude Code: memory (CLAUDE.md) | Imports, lookup order |
| Claude Code: MCP | Scopes, slash commands for MCP prompts |
| API: tool use | Schemas, tool_choice, stop_reason loop |
| API: prompt caching | TTLs and pricing multipliers |
| API: batch processing | 50% discount, when to use |
Practice sources — honest ratings (from people who passed)
| Source | Verdict |
|---|---|
Local questions/ bank (134 Qs) | Best signal. ~30% overlap reported with real exam |
| Official Skilljar practice exam | Good once. Repeating it inflates your score |
| CertSafari | OK for volume. Easier than real. Some wrong answers |
| Udemy exams/courses | Theory foundation only. Much easier than real |
| Community guides (GitHub) | Good topic maps. Contains some errors |
Hands-on beats reading
If you have time: build one small agent with the Agent SDK (tool calls + escalation + structured output). People who passed rate hands-on practice as the biggest score booster.
Time needed
- Using Claude daily already → a few days of prep.
- Otherwise → 1–2 weeks with mocks in exam mode (120-min timer, at least 3 runs).