[{"id":"ccaof-context-behind-instructions","courseId":"ccao-f","domain":"prompting-task-execution","topic":"Explain why behind instructions","prompt":"A comms lead asks Claude to \"keep it short\" for an all-staff update. Drafts come back either long or missing key details. Staff read it on phones during shift breaks, and the safety change must not be lost. Which prompt change best meets the need?","hint":"Claude generalizes better when it understands the purpose of a constraint.","explain":"A length rule can clash with other needs, such as keeping a safety detail. A bare word cap or shouted emphasis gives no basis for deciding what to cut. Consider who reads the output and what it must still contain.","source":"https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#add-context-to-improve-performance","interaction":{"type":"single"},"options":[{"id":"a","text":"Repeat \"KEEP IT SHORT\" in capitals at the start and the end of the prompt.","whyNot":"Louder rules don't tell Claude what to keep, so it may still cut the safety change."},{"id":"b","text":"Explain who reads it, where, and which details must survive the cut.","why":"Claude does better when it knows the reason behind a rule and who the reader is. Explaining the phone-reading setting and the must-keep detail lets it trade length against content sensibly. Shouted or numeric rules alone give it no basis for that trade."},{"id":"c","text":"Set a hard 50-word limit and let Claude decide what to drop.","whyNot":"A cap without priorities invites Claude to drop exactly the detail that matters."},{"id":"d","text":"Ask for three versions every time and always send the shortest one.","whyNot":"Shortest isn't best: nothing checks that the key details survived."}],"answer":"b","explanation":"Claude applies a rule better when it understands its purpose, because the purpose tells it how to handle tradeoffs the rule doesn't spell out.","concept":"Claude applies a rule better when it understands its purpose, because the purpose tells it how to handle tradeoffs the rule doesn't spell out.","rev":1,"kind":"scenario","takeaways":["Give the reason and the audience, not just the rule: Claude generalizes from the why."]},{"id":"ccaof-colleague-test-for-prompts","courseId":"ccao-f","domain":"prompting-task-execution","topic":"Golden rule for clear prompts","prompt":"An operations analyst's prompt \"Summarize the vendor review\" gives uneven results from week to week. Before rewriting it, they want a quick check of whether the prompt is clear enough. Which check best predicts how Claude will handle it?","hint":"Think of Claude as a brilliant new employee who lacks your team's context.","explain":"Writers reading their own draft fill in the background without noticing. A score, a word count or web search each measures something else. Consider which kind of reader would stumble first.","source":"https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#be-clear-and-direct","interaction":{"type":"single"},"options":[{"id":"a","text":"Have a colleague with no background follow it; where they stall, Claude will too.","why":"Anthropic's golden rule is to show the prompt to someone with minimal context and ask them to follow it. Claude is like a brilliant new employee who lacks your norms. Wherever that person would ask questions, the prompt needs more detail."},{"id":"b","text":"Ask Claude to rate the prompt out of 10 and rewrite only if it scores low.","whyNot":"A score doesn't reveal the missing context; a cold reader exposes it directly."},{"id":"c","text":"Count the words: prompts over 100 words are usually unclear.","whyNot":"Length isn't clarity; the vague prompt here is only four words."},{"id":"d","text":"Run it once with web search on and see if the summary improves.","whyNot":"Web search adds outside facts, not the missing instructions about format and scope."}],"answer":"a","explanation":"A prompt only carries what is written in it. Claude doesn't share your background knowledge, so anything you assume but don't state is a gap.","concept":"A prompt only carries what is written in it. Claude doesn't share your background knowledge, so anything you assume but don't state is a gap.","rev":1,"kind":"scenario","takeaways":["Treat Claude like a smart new hire: if a newcomer would need to ask, write it in."]},{"id":"ccaof-decompose-multi-part-request","courseId":"ccao-f","domain":"prompting-task-execution","topic":"Break a complex request down","prompt":"A PM asks Claude in one message to read three stakeholders' interview notes, find conflicting requirements, rank priorities and draft an email to sponsors. The email arrives, but the conflicts are glossed over. What is the most appropriate next step?","hint":"Which step went wrong, and how could you check it before the next one uses it?","explain":"A single request hides its middle steps, so a flaw there reaches the final draft unseen. Consider which output in a multi-part job the rest of the work depends on.","source":"https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#chain-complex-prompts","interaction":{"type":"single"},"options":[{"id":"a","text":"Stage it: confirm the conflicts first, then rank, then draft the email.","why":"Breaking a task into steps lets you inspect intermediate results before they feed the next step. Here the conflict list is the step that failed. Confirming it first makes the ranking and the email trustworthy."},{"id":"b","text":"Send the same request again with \"be thorough and don't skip anything\" added at the end.","whyNot":"A generic nudge doesn't let the PM inspect the conflicts before they shape the email."},{"id":"c","text":"Open a new chat and paste the identical request word for word.","whyNot":"Same request, same structure: the step that failed is still buried mid-task."},{"id":"d","text":"Drop the email from the request so the single prompt is a bit shorter.","whyNot":"The email is still needed; the issue is the unchecked middle step, not length."}],"answer":"a","explanation":"Prompt chaining splits complex work into sequential steps, each with a focused prompt, so errors surface at the step where they happen.","concept":"Prompt chaining splits complex work into sequential steps, each with a focused prompt, so errors surface at the step where they happen.","rev":1,"kind":"scenario","takeaways":["Split multi-step work when you need to check an intermediate result before moving on."]},{"id":"ccaof-few-shot-house-style","courseId":"ccao-f","domain":"prompting-task-execution","topic":"Examples to steer style","prompt":"A marketing coordinator needs product blurbs in the brand's punchy house style. Describing the style in words (\"energetic but not salesy\") gives inconsistent results. Which approach best improves consistency?","hint":"Abstract style words vs concrete samples: how many, and how alike should they be?","explain":"Adjectives such as punchy or warm leave room for many readings. Finished copy carries tone, length and rhythm all at once. Consider what one sample teaches compared with several.","source":"https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#use-examples-effectively","interaction":{"type":"single"},"options":[{"id":"a","text":"Expand the written description with more precise adjectives, such as \"bold, warm, witty, never pushy.\"","whyNot":"More adjectives stay abstract; concrete examples steer tone more reliably."},{"id":"b","text":"Add 3 to 5 varied approved blurbs, marked as examples apart from the instructions.","why":"Examples are one of the most reliable ways to steer format, tone and structure. Anthropic suggests 3 to 5 that are relevant and varied, marked off from the instructions. Variety stops Claude from copying one example's quirks."},{"id":"c","text":"Add one approved blurb and ask Claude to follow its wording closely.","whyNot":"A single example invites copying its phrasing instead of the underlying style."},{"id":"d","text":"Ask Claude to restate the house style before writing each blurb.","whyNot":"Restating a vague description doesn't make it any more concrete."}],"answer":"b","explanation":"Few-shot prompting means including sample inputs or outputs in a prompt; Claude infers patterns from them that are hard to put into words.","concept":"Few-shot prompting means including sample inputs or outputs in a prompt; Claude infers patterns from them that are hard to put into words.","rev":1,"kind":"scenario","takeaways":["Show, don't describe: a few varied examples beat adjectives for tone and format."]},{"id":"ccaof-brainstorm-vs-answer-prompt","courseId":"ccao-f","domain":"prompting-task-execution","topic":"Adapt prompt to brainstorming","prompt":"An education program manager wants fresh ideas for a volunteer thank-you event. They reused a prompt from a report task that asks for \"the single best answer, with sources,\" and the ideas feel narrow. What should they change?","hint":"The reused prompt was built for a different kind of task. What does ideation need?","explain":"Some tasks reward one verified conclusion; others reward a wide spread of options. A required citation or a single-answer demand narrows what comes back. Consider when budget and venue should enter the process.","source":"https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#be-clear-and-direct","interaction":{"type":"single"},"options":[{"id":"a","text":"Ask for many varied ideas, unsourced, then shortlist against budget and venue.","why":"Different tasks need different prompts: analysis wants evidence, brainstorming wants range. Say explicitly what you want, here many varied ideas, and apply constraints as a separate filtering step. Reusing a prompt built for another task type carries its constraints along."},{"id":"b","text":"Keep the prompt as is and turn on research so Claude draws on far more sources for ideas.","whyNot":"More sources keep the one-best-answer framing that is narrowing the ideas."},{"id":"c","text":"Require that every idea cites at least one published study.","whyNot":"A citation rule suits analysis, not open idea generation; it filters ideas out early."},{"id":"d","text":"Keep asking for one best idea, at a higher effort level.","whyNot":"More effort on one answer is still one answer; brainstorming needs range."}],"answer":"a","explanation":"Prompts should match the thinking a task needs: divergent tasks benefit from breadth and loose constraints, convergent ones from precision and evidence.","concept":"Prompts should match the thinking a task needs: divergent tasks benefit from breadth and loose constraints, convergent ones from precision and evidence.","rev":1,"kind":"scenario","takeaways":["Match the prompt to the task type: ask for range when ideating, evidence when analyzing."]}]