Flashcards

Cover the right-hand column and test yourself. Cards are grouped by exam domain.

Domain 1: Prompting and Task Execution

Prompt Answer

Eight elements of a strong prompt Task, context, audience, constraints, format, examples, source material, success criteria

• Most common cause of generic
• output
• Missing context and audience

Prompt chaining Using one step's output as the next step's input Task decomposition Breaking a complex request into ordered, reviewable steps Few-shot prompting Including a few examples of the desired output Why label pasted material To separate instructions from content and contain hidden instructions Best feedback style Specific, actionable, one issue at a time

"Interview me" pattern Ask Claude to list the questions it needs answered before starting Brainstorming strategy Ask for many varied ideas first, then narrow with criteria Positive instruction example "Use sentences under 20 words" rather than "don't be wordy" Grounding instruction "Answer only from the attached source; say if it is not covered" Where repeated corrections belong Project instructions, memory or a Skill

Domain 2: Output Evaluation and Validation

Prompt Answer

Hallucination Fluent, confident output that is false or invented Top verification triggers Numbers, citations, quotes, names, dates, regulated topics, recent events Is Claude's confidence evidence? No

Best check for a citation Open the primary source and find the passage Validating bulk outputs Random sample, measure error rate, fix prompt, re-check When human review is mandatory External, high-stakes, regulated, people-affecting or automated outputs Five-question checklist Accurate, complete, consistent, fair, fit for the reader Completeness failure example Summary omits a section or a material caveat Bias signs Gendered or age-coded language, uneven framing, unjustified assumptions Prompt Answer

Format for spreadsheet data Table or CSV

Format for a reusable, editable document Artifact or created file Comparing drafts Define criteria, compare against them, human makes the final call

Domain 3: Product and Model Selection

• Prompt Answer
• Three model tiers named in the exam guide Haiku, Sonnet, Opus
• Fastest, lowest-cost tier Haiku
• Everyday default tier Sonnet
• Most capable tier Opus
• Model selection rule Smallest tier that meets the quality bar
• Feature for recurring context Projects
• Feature for current, cited, multi-source answers Research mode
• Feature for a few current facts Web search

Feature for exact calculations and files Code execution and file creation When to use extended thinking Complex, multi-step reasoning

Context window Everything Claude can see at once, measured in tokens Three moves for a strained chat Restart, summarize and restart, or persist Incognito chat Not saved to history or memory

Domain 4: Workflow Integration and Solution Design

Prompt Answer

First step in integrating Claude Map the current process

Five task-scoring criteria Volume, text or data heavy, easy to check, low risk, clear rules Augment Speed up a step within the current process

Redesign Change the process because Claude exists

Honest time-saving measure Total time including review and rework Three parts of a stakeholder mes‐

sage

Value with evidence, limits, controls

Where to put human review Where errors are costly, external or irreversible What to escalate API integrations, unreviewed automation, custom connectors, large-scale systems

When to scale a pilot After measured results, working review steps and documentation Common pilot failure causes Poor setup, no shared context, no review step

Appendix e. flashcards

Domain 5: Configuration and Knowledge Management

Prompt Answer

Instructions vs knowledge Instructions say how; knowledge says what Six parts of good Project instructions Purpose, audience and voice, sources, format, boundaries, uncertainty behavior

Most common cause of stale answers Outdated or duplicate knowledge files When to use a connector Source changes often

Connector permissions Same as your account's access

When to use a Skill Procedure reused across many contexts

Memory vs Project instructions Memory for personal preferences everywhere; Project for one body of work Keeping Projects current Named owner and review cadence

• Instructions found inside fetched con‐
• tent
• Treat as data, not commands

Domain 6: Governance, Risk, and Responsible Use

• Prompt Answer
• Four data classes Public, internal, confidential, regulated
• Data minimization Share only what the task needs

Compliant path for restricted identifiers Remove or anonymize, then analyze in an approved tool Is "don't retain this" a control? No

Does incognito make restricted data accept‐

able?

No

Who is accountable for AI-assisted work? The person and organization using it Prompt injection Hidden instructions in content Claude reads

Inappropriate uses Deception, fake reviews, unreviewed people decisions, bypassing controls

When policy is unclear Ask the policy owner before proceeding

Ethics principles Transparency, accountability, fairness, IP respect, security

Domain 7: Troubleshooting and Optimization

• Prompt Answer
• First rule of troubleshooting Diagnose before rewriting
• Why change one thing at a time To know what fixed it

Ignored instruction fix Make it prominent, specific and exemplified Hallucinated product feature fix Ground in a facts sheet, allow "not confirmed," review Slow simple workflow fix Lighter tier, no extended thinking, relevant input only, batching Prompt Answer

Quality dropped over time Check knowledge files and instructions Automation complacency Reviewers relax; counter with structured review and sampling Prompt debugging with Claude Share prompt, output and problem; ask for diagnosis and rewrite

Exam facts

Retake waits 14, 30, then 90 days; at most 4 attempts in 12 months Delivery Pearson VUE, online or test center

Appendix e. flashcards