{"id":19364,"date":"2026-09-15T07:15:12","date_gmt":"2026-09-15T11:15:12","guid":{"rendered":"https:\/\/wp.glbgpt.com\/?p=19364"},"modified":"2026-09-15T07:15:13","modified_gmt":"2026-09-15T11:15:13","slug":"best-claude-prompts","status":"publish","type":"post","link":"https:\/\/wp.glbgpt.com\/kr\/hub\/best-claude-prompts","title":{"rendered":"\uae00\uc4f0\uae30, \uc5f0\uad6c, \ucf54\ub529\uc744 \uc704\ud55c \ucd5c\uace0\uc758 Claude \ud504\ub86c\ud504\ud2b8"},"content":{"rendered":"<p class=\"wp-block-paragraph\">The best Claude prompts make the job easy to understand: give Claude a goal, the material it should use, the constraints that matter, and a way to check the result. For writing, protect the facts and the reader\u2019s next action. For research, connect claims to sources. For coding, define correct behavior and run tests.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Start with the task you need below: draft an email, make sense of a source packet, or fix a Python function. Choose from 12 adaptable templates, then explore three worked examples tested with Claude in GlobalGPT. Each example includes the complete prompt and answer, so you can see what to copy, what to change, and what to check for your own work.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">If your work moves from research to writing and then into code, <a href=\"https:\/\/www.glbgpt.com\/home?inviter=hub_content_home&amp;login=1\">GlobalGPT<\/a> brings Claude and other leading models, along with multiple AI tools, into one affordable subscription and one dashboard. You can keep the broader workflow together and choose the model or function that fits the next step.<\/p>\n\n\n\n<div class=\"wp-block-group is-layout-constrained wp-block-group-is-layout-constrained\">\n<figure class=\"wp-block-image size-large\"><a href=\"https:\/\/www.glbgpt.com\/home\/claude-fable-5-1?inviter=hub_claude51&amp;login=1\"><img alt=\"\" fetchpriority=\"high\" decoding=\"async\" width=\"1024\" height=\"584\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-1024x584.png\" class=\"wp-image-19293\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-1024x584.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-300x171.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-766x437.png 766w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-18x10.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-1536x876.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/02\/claude-fable-5.1-1-2048x1168.png 2048w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/a><\/figure>\n\n\n\n<div class=\"wp-block-buttons is-content-justification-center is-layout-flex wp-container-core-buttons-is-layout-3e41869c wp-block-buttons-is-layout-flex\">\n<div class=\"wp-block-button\"><a class=\"wp-block-button__link has-background wp-element-button\" href=\"https:\/\/www.glbgpt.com\/home\/claude-fable-5-1?inviter=hub_claude51&amp;login=1\" style=\"background:linear-gradient(118deg,rgba(250,183,0,0.79) 0%,rgba(255,106,0,0.8) 100%)\"><strong>\uc9c0\uae08 GlobalGPT\uc5d0\uc11c Claude Fable 5.1\uc744 \uc0ac\uc6a9\ud574 \ubcf4\uc138\uc694<\/strong><\/a><\/div>\n<\/div>\n<\/div>\n\n\n\n<style>.bcp-toc{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-toc *{box-sizing:border-box}.bcp-toc .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-toc p,.bcp-toc li,.bcp-toc summary,.bcp-toc td,.bcp-toc th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-toc p{margin:10px 0}.bcp-toc a{color:#165854;text-decoration:underline}.bcp-toc .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-toc pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-toc code{font:inherit;color:inherit;background:transparent}.bcp-toc details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-toc summary{cursor:pointer;font-weight:650}.bcp-toc .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-toc table{width:100%;border-collapse:collapse;font-size:14px}.bcp-toc th,.bcp-toc td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-toc .bcp-note{font-size:14px;color:#435853}.bcp-toc .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-toc .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-toc{padding:16px;margin:18px 0}.bcp-toc .bcp-title{font-size:19px}.bcp-toc pre{padding:12px;font-size:13px}.bcp-toc th,.bcp-toc td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-toc\"><p class=\"bcp-title\">\uc774 \ud398\uc774\uc9c0\uc5d0\uc11c<\/p><nav aria-label=\"\uae30\uc0ac \ub0b4\uc6a9\"><ol><li><a href=\"#prompt-picker\">Prompts at a glance<\/a><\/li><li><a href=\"#writing-prompts\">Writing prompts<\/a><\/li><li><a href=\"#research-prompts\">Research prompts<\/a><\/li><li><a href=\"#coding-prompts\">Coding prompts<\/a><\/li><li><a href=\"#prompt-tests\">What the tests showed<\/a><\/li><li><a href=\"#improve-prompts\">Fix a weak answer<\/a><\/li><li><a href=\"#faq\">\uc790\uc8fc \ubb3b\ub294 \uc9c8\ubb38<\/a><\/li><\/ol><\/nav><\/div>\n\n\n\n<h2 id=\"prompt-picker\" class=\"wp-block-heading\">Best Claude prompts at a glance<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Choose the template by the decision or deliverable you need. Replace the bracketed fields before sending it. The writing and research templates work with pasted text; coding prompts that ask for file access or execution need a tool that can actually perform those actions.<\/p>\n\n\n\n<style>.bcp-picker a:hover{color:#0e4a42!important;text-decoration-thickness:2px!important}.bcp-picker a:focus-visible{outline:3px solid #c3833e;outline-offset:4px}.bcp-picker .bcp-picker-scroll:focus-visible{outline:3px solid #c3833e;outline-offset:3px}@media(max-width:600px){.bcp-picker th,.bcp-picker td{padding:12px!important}}<\/style><div class=\"bcp-picker\" style=\"box-sizing:border-box;width:100%;max-width:100%;margin:24px 0;border:1px solid #d6e1d9;border-radius:10px;overflow:hidden;background:#fff;font-family:system-ui,-apple-system,sans-serif\"><div class=\"bcp-picker-scroll\" role=\"region\" aria-label=\"Claude prompt quick reference\" tabindex=\"0\" style=\"max-width:100%;overflow-x:auto\"><table aria-label=\"Best Claude prompts at a glance\" style=\"width:100%;min-width:560px;margin:0;border:0;border-collapse:collapse;table-layout:fixed\"><thead><tr><th scope=\"col\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;background:#e8f0e9;font-size:13px;font-weight:700;color:#25483f\">\uc791\uc5c5<\/th><th scope=\"col\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;background:#e8f0e9;font-size:13px;font-weight:700;color:#25483f\">\uc5ec\uae30\uc11c \uc2dc\uc791\ud558\uc138\uc694<\/th><th scope=\"col\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;background:#e8f0e9;font-size:13px;font-weight:700;color:#25483f\">Bring<\/th><\/tr><\/thead><tbody><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e5f1eb;color:#225c47;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uae00\uc4f0\uae30<\/span><a href=\"#email-from-brief\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Write an email from a factual brief<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">W1<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">A factual brief, audience, reply action, and word limit.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e5f1eb;color:#225c47;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uae00\uc4f0\uae30<\/span><a href=\"#match-your-voice\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Match the voice of your own writing<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">W2<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">Two or three samples you wrote and the facts for the new piece.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e5f1eb;color:#225c47;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uae00\uc4f0\uae30<\/span><a href=\"#prioritize-edits\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Find the edits that would matter most<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">W3<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">Your draft, intended audience, purpose, and any passages to preserve.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e5f1eb;color:#225c47;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uae00\uc4f0\uae30<\/span><a href=\"#repurpose-source\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Repurpose one source for two audiences<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">W4<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">One approved source and two audience briefs.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e9edfa;color:#3c5087;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uc5f0\uad6c<\/span><a href=\"#synthesize-sources\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Synthesize sources without flattening disagreements<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">R1<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">Labeled excerpts, their dates, the decision question, and any claims to audit.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e9edfa;color:#3c5087;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uc5f0\uad6c<\/span><a href=\"#audit-citations\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Check a claim against its citation<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">R2<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">The exact claim and the source passage, not just a URL.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e9edfa;color:#3c5087;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uc5f0\uad6c<\/span><a href=\"#compare-options\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Compare options against your decision brief<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">R3<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">Your criteria, constraints, and reliable information about each option.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#e9edfa;color:#3c5087;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\uc5f0\uad6c<\/span><a href=\"#plan-research\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Turn a broad topic into a research plan<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">R4<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">A decision, audience, scope, and deadline.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#fff0df;color:#80511f;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\ucf54\ub529<\/span><a href=\"#debug-from-spec\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Debug against explicit behavior and edge cases<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">C1<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">Code, expected behavior, error or failing case, runtime, and constraints.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#fff0df;color:#80511f;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\ucf54\ub529<\/span><a href=\"#tests-from-spec\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Write tests from a specification<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">C2<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">Behavioral requirements and the project test framework.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#fff0df;color:#80511f;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\ucf54\ub529<\/span><a href=\"#narrow-refactor\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Refactor without expanding the job<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">C3<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">The code, the specific maintenance problem, and compatibility requirements.<\/td><\/tr><tr style=\"background:#fff\"><th scope=\"row\" style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;font-weight:400;width:38%\"><span style=\"display:block;width:fit-content;margin:0 0 7px;padding:3px 8px;border-radius:4px;background:#fff0df;color:#80511f;font-size:11px;font-weight:700;letter-spacing:.04em;line-height:1.5\">\ucf54\ub529<\/span><a href=\"#explain-code\" style=\"color:#17645b;font-weight:650;text-decoration:underline;text-underline-offset:3px\">Explain unfamiliar code through a concrete input<\/a><\/th><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:16%\"><span style=\"display:inline-block;padding:4px 9px;border:1px solid #d8e3dc;border-radius:5px;background:#f5f8f5;color:#34554b;font:600 13px\/1.5 ui-monospace,monospace\">C4<\/span><\/td><td style=\"padding:16px 18px;border:0;border-bottom:1px solid #e0e7e2;text-align:left;vertical-align:top;line-height:1.55;font-size:15px;color:#233d39;width:46%\">The actual code and one representative input.<\/td><\/tr><\/tbody><\/table><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">A reusable starting format is enough for many jobs. <a href=\"https:\/\/platform.claude.com\/docs\/en\/build-with-claude\/prompt-engineering\/overview\">Anthropic\u2019s prompt engineering overview<\/a> begins with success criteria and a way to test them. You do not need an elaborate expert persona to say what a useful answer should contain.<\/p>\n\n\n\n<style>.bcp-starter{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-starter *{box-sizing:border-box}.bcp-starter .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-starter p,.bcp-starter li,.bcp-starter summary,.bcp-starter td,.bcp-starter th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-starter p{margin:10px 0}.bcp-starter a{color:#165854;text-decoration:underline}.bcp-starter .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-starter pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-starter code{font:inherit;color:inherit;background:transparent}.bcp-starter details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-starter summary{cursor:pointer;font-weight:650}.bcp-starter .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-starter table{width:100%;border-collapse:collapse;font-size:14px}.bcp-starter th,.bcp-starter td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-starter .bcp-note{font-size:14px;color:#435853}.bcp-starter .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-starter .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-starter{padding:16px;margin:18px 0}.bcp-starter .bcp-title{font-size:19px}.bcp-starter pre{padding:12px;font-size:13px}.bcp-starter th,.bcp-starter td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-starter\"><p class=\"bcp-title\">A short starting template<\/p><pre><code>Help me [TASK] for [AUDIENCE OR PURPOSE].\nUse this material: [PASTE MATERIAL].\nRespect these constraints: [FACTS, SCOPE, LENGTH, OR RULES].\nReturn [OUTPUT FORMAT].\nA successful result must [CHECKABLE REQUIREMENTS].\nIf a necessary detail is missing, ask the specific question that blocks the task.<\/code><\/pre><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Use headings or tags when they make a complex brief easier to separate. <a href=\"https:\/\/platform.claude.com\/docs\/en\/build-with-claude\/prompt-engineering\/claude-prompting-best-practices\">Anthropic\u2019s current prompting guidance<\/a> recommends clear instructions, useful examples, and descriptive XML tags for mixed material. Tags organize the text; they do not create evidence, grant tool access, or make an answer correct.<\/p>\n\n\n\n<h2 id=\"writing-prompts\" class=\"wp-block-heading\">Claude prompts for writing<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">For writing, the hard part is often preserving the brief while changing the expression. State what the reader needs to do and which details are fixed. A smooth sentence is not helpful if it quietly adds a promise.<\/p>\n\n\n\n<h3 id=\"email-from-brief\" class=\"wp-block-heading\">1. Write an email from a factual brief<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A clear invitation, update, or customer email with details that must stay exact. <strong>Bring:<\/strong> A factual brief, audience, reply action, and word limit.<\/p>\n\n\n\n<style>.bcp-prompt-w1{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-w1 *{box-sizing:border-box}.bcp-prompt-w1 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-w1 p,.bcp-prompt-w1 li,.bcp-prompt-w1 summary,.bcp-prompt-w1 td,.bcp-prompt-w1 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-w1 p{margin:10px 0}.bcp-prompt-w1 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-w1 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-w1 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-w1 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-w1 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-w1 summary{cursor:pointer;font-weight:650}.bcp-prompt-w1 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-w1 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-w1 th,.bcp-prompt-w1 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-w1 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-w1 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-w1 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-w1{padding:16px;margin:18px 0}.bcp-prompt-w1 .bcp-title{font-size:19px}.bcp-prompt-w1 pre{padding:12px;font-size:13px}.bcp-prompt-w1 th,.bcp-prompt-w1 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-w1\"><p class=\"bcp-kicker\">W1 \u00b7 Representative task tested below<\/p><pre><code>Write the email requested in the brief. Treat confirmed facts as fixed. Keep requests distinct from confirmations. Follow the stated output format and length. Use only the supplied details.\n\n&lt;brief&gt;\nAudience: [WHO WILL READ THIS]\nPurpose: [WHAT THE EMAIL SHOULD ACHIEVE]\nConfirmed facts: [DATES, TIMES, PRICES, LIMITS, AND OTHER DETAILS]\nUnknown or unapproved details: [WHAT YOU CANNOT PROMISE]\nReader action: [EXACT NEXT STEP]\nTone: [TONE]\nOutput: a subject line and [WORD RANGE]-word body.\n&lt;\/brief&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Check every date, number, promise, and next action against the brief. A friendly extra sentence can still create an obligation.<\/p><\/div>\n\n\n\n<h3 id=\"match-your-voice\" class=\"wp-block-heading\">2. Match the voice of your own writing<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A short draft that sounds consistent with your existing work. <strong>Bring:<\/strong> Two or three samples you wrote and the facts for the new piece.<\/p>\n\n\n\n<style>.bcp-prompt-w2{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-w2 *{box-sizing:border-box}.bcp-prompt-w2 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-w2 p,.bcp-prompt-w2 li,.bcp-prompt-w2 summary,.bcp-prompt-w2 td,.bcp-prompt-w2 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-w2 p{margin:10px 0}.bcp-prompt-w2 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-w2 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-w2 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-w2 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-w2 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-w2 summary{cursor:pointer;font-weight:650}.bcp-prompt-w2 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-w2 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-w2 th,.bcp-prompt-w2 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-w2 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-w2 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-w2 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-w2{padding:16px;margin:18px 0}.bcp-prompt-w2 .bcp-title{font-size:19px}.bcp-prompt-w2 pre{padding:12px;font-size:13px}.bcp-prompt-w2 th,.bcp-prompt-w2 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-w2\"><p class=\"bcp-kicker\">W2 \u00b7 Adaptable template<\/p><pre><code>Use my writing samples to identify three observable style traits: sentence rhythm, level of formality, and how I explain ideas. Apply those traits to a new [FORMAT] for [AUDIENCE].\n\nUse the samples for style only. Use the new brief for facts. Give the three traits, then the draft in [WORD RANGE] words.\n\n&lt;my_samples&gt;\n[PASTE YOUR OWN WRITING]\n&lt;\/my_samples&gt;\n&lt;new_brief&gt;\n[TOPIC, PURPOSE, CONFIRMED FACTS, AND READER ACTION]\n&lt;\/new_brief&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Look for sentence rhythm and specificity, then check that sample facts have not leaked into the new draft.<\/p><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Choose samples whose style you want to retain. A list of adjectives such as \u201cfriendly, insightful, professional\u201d leaves more room for interpretation than a few concrete examples. The new brief should remain the source of factual details.<\/p>\n\n\n\n<h3 id=\"prioritize-edits\" class=\"wp-block-heading\">3. Find the edits that would matter most<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A draft that feels weak, without inviting a complete rewrite. <strong>Bring:<\/strong> Your draft, intended audience, purpose, and any passages to preserve.<\/p>\n\n\n\n<style>.bcp-prompt-w3{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-w3 *{box-sizing:border-box}.bcp-prompt-w3 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-w3 p,.bcp-prompt-w3 li,.bcp-prompt-w3 summary,.bcp-prompt-w3 td,.bcp-prompt-w3 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-w3 p{margin:10px 0}.bcp-prompt-w3 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-w3 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-w3 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-w3 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-w3 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-w3 summary{cursor:pointer;font-weight:650}.bcp-prompt-w3 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-w3 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-w3 th,.bcp-prompt-w3 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-w3 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-w3 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-w3 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-w3{padding:16px;margin:18px 0}.bcp-prompt-w3 .bcp-title{font-size:19px}.bcp-prompt-w3 pre{padding:12px;font-size:13px}.bcp-prompt-w3 th,.bcp-prompt-w3 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-w3\"><p class=\"bcp-kicker\">W3 \u00b7 Adaptable template<\/p><pre><code>Review this draft for [AUDIENCE] who need to [READER GOAL]. Identify up to three changes that would most improve clarity or usefulness. If fewer than three are needed, say so.\n\nFor each change, quote the original passage, explain the reader problem, and propose a replacement. Preserve the factual meaning and these elements: [MUST KEEP]. Do not rewrite the entire draft.\n\n&lt;draft&gt;\n[PASTE DRAFT]\n&lt;\/draft&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Reject findings that cannot point to a real passage or a meaningful reader problem.<\/p><\/div>\n\n\n\n<h3 id=\"repurpose-source\" class=\"wp-block-heading\">4. Repurpose one source for two audiences<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Turning an approved update into an internal note and a customer message. <strong>Bring:<\/strong> One approved source and two audience briefs.<\/p>\n\n\n\n<style>.bcp-prompt-w4{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-w4 *{box-sizing:border-box}.bcp-prompt-w4 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-w4 p,.bcp-prompt-w4 li,.bcp-prompt-w4 summary,.bcp-prompt-w4 td,.bcp-prompt-w4 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-w4 p{margin:10px 0}.bcp-prompt-w4 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-w4 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-w4 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-w4 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-w4 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-w4 summary{cursor:pointer;font-weight:650}.bcp-prompt-w4 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-w4 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-w4 th,.bcp-prompt-w4 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-w4 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-w4 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-w4 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-w4{padding:16px;margin:18px 0}.bcp-prompt-w4 .bcp-title{font-size:19px}.bcp-prompt-w4 pre{padding:12px;font-size:13px}.bcp-prompt-w4 th,.bcp-prompt-w4 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-w4\"><p class=\"bcp-kicker\">W4 \u00b7 Adaptable template<\/p><pre><code>Adapt the approved source into two short versions:\nA. [FORMAT, AUDIENCE, PURPOSE, WORD LIMIT]\nB. [FORMAT, AUDIENCE, PURPOSE, WORD LIMIT]\n\nKeep the dates, numbers, feature availability, and commitments consistent. Change the explanation and emphasis to suit each audience. If an audience needs a detail the source does not contain, list that gap after the drafts.\n\n&lt;approved_source&gt;\n[PASTE SOURCE]\n&lt;\/approved_source&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Compare both versions with the same fact list; a new audience does not authorize new features or promises.<\/p><\/div>\n\n\n\n<style>.bcp-method{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-method *{box-sizing:border-box}.bcp-method .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-method p,.bcp-method li,.bcp-method summary,.bcp-method td,.bcp-method th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-method p{margin:10px 0}.bcp-method a{color:#165854;text-decoration:underline}.bcp-method .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-method pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-method code{font:inherit;color:inherit;background:transparent}.bcp-method details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-method summary{cursor:pointer;font-weight:650}.bcp-method .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-method table{width:100%;border-collapse:collapse;font-size:14px}.bcp-method th,.bcp-method td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-method .bcp-note{font-size:14px;color:#435853}.bcp-method .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-method .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-method{padding:16px;margin:18px 0}.bcp-method .bcp-title{font-size:19px}.bcp-method pre{padding:12px;font-size:13px}.bcp-method th,.bcp-method td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-method\"><p class=\"bcp-kicker\">Try these tasks in GlobalGPT<\/p><p class=\"bcp-title\">Turn a brief into an email, a source packet into a recommendation, and a bug report into a fix.<\/p><p>We used Claude Sonnet 5 in GlobalGPT for all three tasks. The email drafts retained the key event details, the research answers identified reasons to hold off on a rollout, and both Python fixes passed all 24 checks when we ran them separately. These are useful starting points for the work you can bring into the same dashboard.<\/p><p>To try it yourself, choose a template below and replace the sample material with your own. GlobalGPT brings writing, research, coding, and other AI tools together in one affordable subscription, so you can move from understanding a problem to drafting the next deliverable without juggling separate services.<\/p><details><summary>How we compared the prompts<\/summary><p>On September 14, 2026, each task received two prompts with the same fictional source material and requirements: one brief instruction and one more structured version. We kept the first complete answer from each, for six answers in total. Research used the supplied passages; the returned Python code was checked separately, not executed by the model during the conversation.<\/p><p>These examples cover three tasks, once per version. The complete inputs and answers appear below, along with the edits or checks each result needs. The 12 reusable templates have not each been tested separately.<\/p><\/details><\/div>\n\n\n\n<style>.bcp-test-w{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-test-w *{box-sizing:border-box}.bcp-test-w .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-test-w p,.bcp-test-w li,.bcp-test-w summary,.bcp-test-w td,.bcp-test-w th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-test-w p{margin:10px 0}.bcp-test-w a{color:#165854;text-decoration:underline}.bcp-test-w .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-test-w pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-test-w code{font:inherit;color:inherit;background:transparent}.bcp-test-w details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-test-w summary{cursor:pointer;font-weight:650}.bcp-test-w .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-test-w table{width:100%;border-collapse:collapse;font-size:14px}.bcp-test-w th,.bcp-test-w td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-test-w .bcp-note{font-size:14px;color:#435853}.bcp-test-w .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-test-w .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-test-w{padding:16px;margin:18px 0}.bcp-test-w .bcp-title{font-size:19px}.bcp-test-w pre{padding:12px;font-size:13px}.bcp-test-w th,.bcp-test-w td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-test-w\"><p class=\"bcp-kicker\">First-output comparison \u00b7 W<\/p><p class=\"bcp-title\">Draft an event email with the key details in place.<\/p><div class=\"bcp-scroll\"><table><thead><tr><th scope=\"col\">\ud655\uc778<\/th><th scope=\"col\">Plain instruction<\/th><th scope=\"col\">Structured instruction<\/th><\/tr><\/thead><tbody><tr><td>Core facts and action checks<\/td><td>8 \/ 9<\/td><td>8 \/ 9<\/td><\/tr><tr><td>Subject and body length<\/td><td>Pass \u00b7 126 words<\/td><td>Pass \u00b7 128 words<\/td><\/tr><tr><td>Unapproved follow-up promise<\/td><td>\ud604\uc7ac<\/td><td>\ud604\uc7ac<\/td><\/tr><\/tbody><\/table><\/div><p>Both versions kept the date, times, capacity, price, notes, and the distinction between requesting and confirming a booking. That gives you a draft to edit rather than a blank page. Before sending, remove the follow-up promise that neither brief authorized; the plain version also called the sessions \u201cidentical\u201d without support. For your own email, list both the facts to keep and the commitments you are allowed to make.<\/p><p class=\"bcp-note\">The eight passing checks cover the date; both times and UTC; duration; per-session capacity; free attendance; downloadable notes; no recording; and requesting rather than confirming a booking. The ninth checks for invented facts or commitments. Word counts exclude the subject.<\/p><p class=\"bcp-note\">Elapsed time: 9.73s \/ 7.02s. Reported input + output tokens: 400 + 617 \/ 468 + 299. One response each; no retries.<\/p><details><summary>Plain instruction \u2014 complete prompt and source material<\/summary><pre><code>Write the email requested in the brief below.\n\nFICTIONAL TEST FIXTURE. These are invented workshop details, not a real offer.\n\nYou are writing an invitation email for a fictional community workshop named Clear Notes.\nAudience: small-team project leads who want cleaner meeting notes.\nGoal: invite the reader to reply MORNING or AFTERNOON to request a place. A reply requests a place; it does not confirm a booking.\nConfirmed facts:\n- Date: 8 October 2026.\n- Two separate sessions start at 09:00 UTC and 16:00 UTC.\n- Each session lasts 40 minutes.\n- Each session has 25 places.\n- Attendance is free.\n- Attendees receive downloadable notes.\n- No recording will be provided.\n- No registration deadline has been set.\nDeliverable: a subject line and an email body of 100\u2013140 words. Plain, friendly English; no hype. Include the date, both start times and timezone, duration, per-session capacity, price, downloadable notes, lack of recording, and reply action. Do not invent a link, deadline, confirmed reservation, speaker, or venue. It is fine to leave the unset registration deadline unmentioned.\n<\/code><\/pre><\/details><details><summary>Plain instruction \u2014 complete original answer<\/summary><pre><code>**Subject: Clear Notes Workshop \u2013 8 October 2026 \u2013 Reply to Request Your Place**\n\nHello,\n\nWe&#x27;re inviting project leads who want cleaner, more useful meeting notes to join Clear Notes, a free community workshop on 8 October 2026.\n\nThere are two identical sessions, each lasting 40 minutes: one starting at 09:00 UTC and one at 16:00 UTC. Each session has 25 places available.\n\nAttendance is free, and everyone who takes part will receive downloadable notes to keep. Please note that no recording will be provided, so live attendance is the only way to take part.\n\nTo request a place, simply reply to this email with MORNING or AFTERNOON, depending on which session suits you best. Replying requests a place; it does not confirm your booking. We&#x27;ll follow up with more details soon.\n\nThanks, and we hope to see you there.<\/code><\/pre><\/details><details><summary>Structured instruction \u2014 complete prompt and source material<\/summary><pre><code>Write the invitation email requested in the brief. Treat confirmed facts as fixed. Keep a request for a place distinct from a confirmed booking. Follow the stated output format and length. Use only the supplied details.\n\n&lt;task_material&gt;\nFICTIONAL TEST FIXTURE. These are invented workshop details, not a real offer.\n\nYou are writing an invitation email for a fictional community workshop named Clear Notes.\nAudience: small-team project leads who want cleaner meeting notes.\nGoal: invite the reader to reply MORNING or AFTERNOON to request a place. A reply requests a place; it does not confirm a booking.\nConfirmed facts:\n- Date: 8 October 2026.\n- Two separate sessions start at 09:00 UTC and 16:00 UTC.\n- Each session lasts 40 minutes.\n- Each session has 25 places.\n- Attendance is free.\n- Attendees receive downloadable notes.\n- No recording will be provided.\n- No registration deadline has been set.\nDeliverable: a subject line and an email body of 100\u2013140 words. Plain, friendly English; no hype. Include the date, both start times and timezone, duration, per-session capacity, price, downloadable notes, lack of recording, and reply action. Do not invent a link, deadline, confirmed reservation, speaker, or venue. It is fine to leave the unset registration deadline unmentioned.\n&lt;\/task_material&gt;\n<\/code><\/pre><\/details><details><summary>Structured instruction \u2014 complete original answer<\/summary><pre><code>Subject: You&#x27;re Invited: Clear Notes Workshop \u2013 8 October 2026\n\nHi there,\n\nIf you&#x27;re a project lead looking for cleaner meeting notes, we&#x27;d love to have you at Clear Notes on 8 October 2026.\n\nTwo sessions are available, each lasting 40 minutes: one starting at 09:00 UTC and one at 16:00 UTC. Each session holds 25 places, and attendance is completely free.\n\nAll attendees will receive downloadable notes to keep and reference afterward. Please note that no recording will be provided, so live attendance is the only way to take part.\n\nTo request a place, simply reply with MORNING or AFTERNOON, depending on which session works best for you. Replying requests a place; it does not confirm your booking; we&#x27;ll follow up with further details soon.\n\nWe hope to see you there.\n\nWarm regards,\nThe Clear Notes Team<\/code><\/pre><\/details><\/div>\n\n\n\n<h2 id=\"research-prompts\" class=\"wp-block-heading\">Claude prompts for research<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Research prompts need an evidence boundary. Supply the relevant passages with source IDs, or use a tool that can actually retrieve them. A URL alone is not proof that Claude has opened a page. These templates separate what the sources say from what you can conclude.<\/p>\n\n\n\n<h3 id=\"synthesize-sources\" class=\"wp-block-heading\">5. Synthesize sources without flattening disagreements<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Answering a decision question from documents you can inspect. <strong>Bring:<\/strong> Labeled excerpts, their dates, the decision question, and any claims to audit.<\/p>\n\n\n\n<style>.bcp-prompt-r1{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-r1 *{box-sizing:border-box}.bcp-prompt-r1 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-r1 p,.bcp-prompt-r1 li,.bcp-prompt-r1 summary,.bcp-prompt-r1 td,.bcp-prompt-r1 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-r1 p{margin:10px 0}.bcp-prompt-r1 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-r1 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-r1 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-r1 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-r1 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-r1 summary{cursor:pointer;font-weight:650}.bcp-prompt-r1 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-r1 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-r1 th,.bcp-prompt-r1 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-r1 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-r1 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-r1 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-r1{padding:16px;margin:18px 0}.bcp-prompt-r1 .bcp-title{font-size:19px}.bcp-prompt-r1 pre{padding:12px;font-size:13px}.bcp-prompt-r1 th,.bcp-prompt-r1 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-r1\"><p class=\"bcp-kicker\">R1 \u00b7 Representative task tested below<\/p><pre><code>Answer [DECISION QUESTION] using only the source packet below. Compare definitions, populations, and denominators before treating numerical differences as evidence of improvement. Distinguish a draft assertion from measurement. Attach each conclusion to an exact source quote that actually supports it, and leave absent evidence absent.\n\nOutput: a recommendation of at most [LIMIT] words; a table with Claim, Verdict, Source ID, and Exact supporting quote; and three missing measurements needed for a fair comparison. Use Supported, Partly supported, Unsupported, or Contradicted only when the evidence warrants that label. Audit these claims: [CLAIMS].\n\n&lt;source_packet&gt;\n[SOURCE IDS, DATES, AND COMPLETE RELEVANT EXCERPTS]\n&lt;\/source_packet&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Check quoted text, the claim it supports, and whether quantities use the same definition and denominator.<\/p><\/div>\n\n\n\n<h3 id=\"audit-citations\" class=\"wp-block-heading\">6. Check a claim against its citation<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Reviewing a draft before a citation is used as evidence. <strong>Bring:<\/strong> The exact claim and the source passage, not just a URL.<\/p>\n\n\n\n<style>.bcp-prompt-r2{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-r2 *{box-sizing:border-box}.bcp-prompt-r2 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-r2 p,.bcp-prompt-r2 li,.bcp-prompt-r2 summary,.bcp-prompt-r2 td,.bcp-prompt-r2 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-r2 p{margin:10px 0}.bcp-prompt-r2 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-r2 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-r2 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-r2 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-r2 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-r2 summary{cursor:pointer;font-weight:650}.bcp-prompt-r2 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-r2 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-r2 th,.bcp-prompt-r2 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-r2 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-r2 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-r2 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-r2{padding:16px;margin:18px 0}.bcp-prompt-r2 .bcp-title{font-size:19px}.bcp-prompt-r2 pre{padding:12px;font-size:13px}.bcp-prompt-r2 th,.bcp-prompt-r2 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-r2\"><p class=\"bcp-kicker\">R2 \u00b7 Adaptable template<\/p><pre><code>Check each claim against the source text supplied with it. For each, return: claim, source ID, exact relevant quote, verdict, and a corrected version if needed.\n\nUse Supported when the source backs the whole claim; Partly supported when it backs only part; Unsupported when evidence is absent or insufficient; and Contradicted when the source directly conflicts with the claim. If the source text is missing, say you cannot assess support. Do not infer the source contents from its title or URL.\n\n[PASTE CLAIM\u2013SOURCE PAIRS]<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> A real source can still fail to support the claim. Absence of evidence is not a contradiction.<\/p><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">\u201cUnsupported\u201d means the supplied evidence does not establish the claim. \u201cContradicted\u201d means the evidence conflicts with it. That distinction matters when a source reports no results yet: it has not shown either improvement or deterioration.<\/p>\n\n\n\n<h3 id=\"compare-options\" class=\"wp-block-heading\">7. Compare options against your decision brief<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Choosing among options with uneven evidence. <strong>Bring:<\/strong> Your criteria, constraints, and reliable information about each option.<\/p>\n\n\n\n<style>.bcp-prompt-r3{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-r3 *{box-sizing:border-box}.bcp-prompt-r3 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-r3 p,.bcp-prompt-r3 li,.bcp-prompt-r3 summary,.bcp-prompt-r3 td,.bcp-prompt-r3 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-r3 p{margin:10px 0}.bcp-prompt-r3 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-r3 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-r3 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-r3 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-r3 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-r3 summary{cursor:pointer;font-weight:650}.bcp-prompt-r3 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-r3 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-r3 th,.bcp-prompt-r3 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-r3 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-r3 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-r3 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-r3{padding:16px;margin:18px 0}.bcp-prompt-r3 .bcp-title{font-size:19px}.bcp-prompt-r3 pre{padding:12px;font-size:13px}.bcp-prompt-r3 th,.bcp-prompt-r3 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-r3\"><p class=\"bcp-kicker\">R3 \u00b7 Adaptable template<\/p><pre><code>Compare [OPTIONS] for [DECISION]. Use these criteria in this priority order: [CRITERIA]. Hard constraints: [CONSTRAINTS]. Use only the option information below.\n\nReturn a comparison table, the best-supported choice, and the missing fact most likely to change that choice. Mark missing values as unknown. Keep facts separate from judgments. Do not invent prices, scores, or numerical weights. If no option can be selected from the evidence, explain the specific unresolved trade-off.\n\n&lt;option_information&gt;\n[PASTE VERIFIED INFORMATION WITH SOURCE IDS]\n&lt;\/option_information&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Check whether a recommendation depends on an unknown value or on weights you never chose.<\/p><\/div>\n\n\n\n<h3 id=\"plan-research\" class=\"wp-block-heading\">8. Turn a broad topic into a research plan<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Starting research before you have enough sources to answer. <strong>Bring:<\/strong> A decision, audience, scope, and deadline.<\/p>\n\n\n\n<style>.bcp-prompt-r4{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-r4 *{box-sizing:border-box}.bcp-prompt-r4 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-r4 p,.bcp-prompt-r4 li,.bcp-prompt-r4 summary,.bcp-prompt-r4 td,.bcp-prompt-r4 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-r4 p{margin:10px 0}.bcp-prompt-r4 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-r4 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-r4 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-r4 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-r4 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-r4 summary{cursor:pointer;font-weight:650}.bcp-prompt-r4 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-r4 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-r4 th,.bcp-prompt-r4 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-r4 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-r4 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-r4 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-r4{padding:16px;margin:18px 0}.bcp-prompt-r4 .bcp-title{font-size:19px}.bcp-prompt-r4 pre{padding:12px;font-size:13px}.bcp-prompt-r4 th,.bcp-prompt-r4 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-r4\"><p class=\"bcp-kicker\">R4 \u00b7 Adaptable template<\/p><pre><code>Build a research plan for [QUESTION] to help [AUDIENCE] decide [DECISION]. Scope: [REGION, PERIOD, AND EXCLUSIONS]. Time available: [TIME].\n\nReturn five focused subquestions. For each, give the evidence needed, a preferred primary-source type, suggested search terms, and a finding that would change the decision. End with the two highest-priority searches. Do not invent studies or claim that you have read sources. If you use browsing tools, separately identify the pages actually opened and what each supports.<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> The output is a plan. Treat proposed searches and potential sources as leads, not completed research.<\/p><\/div>\n\n\n\n<style>.bcp-test-r{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-test-r *{box-sizing:border-box}.bcp-test-r .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-test-r p,.bcp-test-r li,.bcp-test-r summary,.bcp-test-r td,.bcp-test-r th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-test-r p{margin:10px 0}.bcp-test-r a{color:#165854;text-decoration:underline}.bcp-test-r .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-test-r pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-test-r code{font:inherit;color:inherit;background:transparent}.bcp-test-r details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-test-r summary{cursor:pointer;font-weight:650}.bcp-test-r .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-test-r table{width:100%;border-collapse:collapse;font-size:14px}.bcp-test-r th,.bcp-test-r td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-test-r .bcp-note{font-size:14px;color:#435853}.bcp-test-r .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-test-r .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-test-r{padding:16px;margin:18px 0}.bcp-test-r .bcp-title{font-size:19px}.bcp-test-r pre{padding:12px;font-size:13px}.bcp-test-r th,.bcp-test-r td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-test-r\"><p class=\"bcp-kicker\">First-output comparison \u00b7 R<\/p><p class=\"bcp-title\">A missing result is not a disproved claim.<\/p><div class=\"bcp-scroll\"><table><thead><tr><th scope=\"col\">\ud655\uc778<\/th><th scope=\"col\">Plain instruction<\/th><th scope=\"col\">Structured instruction<\/th><\/tr><\/thead><tbody><tr><td>Predefined research checks<\/td><td>6 \/ 8<\/td><td>7 \/ 8<\/td><\/tr><tr><td>Retention evidence label<\/td><td>Contradicted \u00b7 too strong<\/td><td>Unsupported \u00b7 appropriate<\/td><\/tr><tr><td>Invited accounts vs onboarding starts<\/td><td>Not explicitly explained<\/td><td>Not explicitly explained<\/td><\/tr><\/tbody><\/table><\/div><p>Both answers rejected a full rollout and recognized that the pilots used different activation definitions and populations. The structured version correctly labeled the retention claim \u201cUnsupported.\u201d The plain version wrote \u201cContradicted,\u201d even though its quote only established that no retention results existed. Neither answer explicitly highlighted the different denominators.<\/p><p>The rates were 40\/50 invited accounts (80%) and 54\/60 onboarding starts (90%). The arithmetic is correct; the measurements are not interchangeable. The structured answer added a separate check of whether \u201cactivation\u201d meant the same thing. That was an extra analytical question, not a fourth claim made by the rollout memo.<\/p><div class=\"bcp-scroll\"><table><thead><tr><th scope=\"col\">Memo claim<\/th><th scope=\"col\">What the supplied source supports<\/th><th scope=\"col\">Appropriate treatment<\/th><\/tr><\/thead><tbody><tr><td>B raised activation by 10 percentage points<\/td><td>Different rates, definitions, denominators, and populations<\/td><td>Do not treat the gap as a causal improvement<\/td><\/tr><tr><td>B improved retention<\/td><td>S3: no 30-day retention results yet<\/td><td>Unsupported; not disproved<\/td><\/tr><tr><td>B reduced support time to six hours<\/td><td>S3: pooled median, no pilot breakdown<\/td><td>No demonstrated reduction<\/td><\/tr><\/tbody><\/table><\/div><details><summary>See all eight research checks<\/summary><pre><code>1. A: 40\/50 = 80%, using invited accounts. Both pass.\n2. B: 54\/60 = 90%, using onboarding starts. Both pass; source quotes contain the denominators.\n3. Explicitly flag both different activation definitions and different denominators. Both fail the denominator part.\n4. Do not treat absent retention results as proof of improvement. Both explain the data gap.\n5. Do not treat a pooled support median as a demonstrated decrease. Both pass.\n6. Do not infer a causal benefit or justify a full rollout from these pilots. Both pass.\n7. Match each claim and evidence label to what its cited quote supports. Plain fails on Contradicted; structured passes.\n8. Identify three specific missing measurements. Both pass.\nThe plain answer\u2019s incorrect label is counted once, in check 7.<\/code><\/pre><\/details><p class=\"bcp-note\">Elapsed time: 31.35s \/ 31.07s. Reported input + output tokens: 743 + 2822 \/ 832 + 2837. One response each; no retries.<\/p><details><summary>Plain instruction \u2014 complete prompt and source material<\/summary><pre><code>Answer the assignment at the end of this source packet.\n\n# Fictional research packet\n\nAll organizations, data, quotations, and documents below are invented for a reproducible test. Source IDs refer only to this packet; they are not public studies.\n\n## S1 \u2014 Pilot A closeout, 2 September 2026\n\n&quot;Pilot A invited 50 accounts; 40 completed activation within seven days.&quot;\n&quot;Activation meant creating a workspace and inviting at least one teammate.&quot;\n&quot;Accounts were existing customers selected by the customer success team.&quot;\n&quot;The pilot used guided onboarding.&quot;\n\n## S2 \u2014 Pilot B dashboard note, 3 September 2026\n\n&quot;Pilot B recorded 60 onboarding starts; 54 accounts created a workspace within seven days.&quot;\n&quot;This dashboard calls workspace creation activation; it does not require a teammate invitation.&quot;\n&quot;The invitation count was not captured.&quot;\n&quot;Accounts were new self-serve signups using a checklist.&quot;\n\n## S3 \u2014 Analyst review, 4 September 2026\n\n&quot;The pilots were not randomized and used different account populations.&quot;\n&quot;The median support response time was six hours across the two pilots combined; it was not broken out by pilot.&quot;\n&quot;No 30-day retention or revenue results are available yet.&quot;\n\n## S4 \u2014 Draft rollout memo, 5 September 2026\n\n&quot;Pilot B raised activation by 10 percentage points, improved retention, and reduced support time to six hours. Roll out the checklist to every account.&quot;\nThis is a proposal draft, not an additional data source. Its assertions must be checked against S1\u2013S3.\n\n## Assignment shared by both prompt versions\n\nAnswer: Does this packet justify rolling out Pilot B to all accounts?\nUse only S1\u2013S4. Output (1) a recommendation of at most 120 words, (2) a table with columns Claim, Verdict, Source ID, Exact supporting quote, and (3) three missing measurements needed for a fair comparison.\nAudit all three empirical claims in the rollout memo: the activation increase, improved retention, and reduced support time. Use verdicts Supported, Partly supported, Unsupported, or Contradicted, with a short explanation if needed. Include the numerical rates and their denominators in the recommendation or table. Do not invent citations, URLs, causal effects, or findings absent from the packet.\n<\/code><\/pre><\/details><details><summary>Plain instruction \u2014 complete original answer<\/summary><pre><code>## Recommendation (\u2264120 words)\n\nDo not roll out Pilot B yet. Pilot B&#x27;s raw activation rate (54\/60 = 90%) is 10 percentage points higher than Pilot A&#x27;s (40\/50 = 80%), but the two pilots measured &quot;activation&quot; differently (B does not require a teammate invite), drew from different populations (new self-serve signups vs. existing customers), and were not randomized (S3) \u2014 so the gap cannot be attributed to the checklist. The retention claim is unsupported: S3 explicitly states no 30-day retention data exists yet. The &quot;six hours&quot; support-time figure is a combined median across both pilots, not broken out by pilot, so no reduction specific to Pilot B is demonstrated. A controlled, matched, consistently-defined comparison is needed before rollout.\n\n## Claim audit table\n\n| Claim | Verdict | Source ID | Exact supporting quote |\n|---|---|---|---|\n| &quot;Pilot B raised activation by 10 percentage points&quot; | Partly supported \u2014 raw rates differ by 10pp (54\/60=90% vs 40\/50=80%), but definitions\/populations\/design differ, so the difference is not attributable to the checklist | S1 | &quot;Pilot A invited 50 accounts; 40 completed activation within seven days.&quot; |\n| (same claim, cont.) | | S2 | &quot;Pilot B recorded 60 onboarding starts; 54 accounts created a workspace within seven days.&quot; \/ &quot;it does not require a teammate invitation&quot; |\n| (same claim, cont.) | | S3 | &quot;The pilots were not randomized and used different account populations.&quot; |\n| &quot;improved retention&quot; | Contradicted \u2014 no retention data exists to support this | S3 | &quot;No 30-day retention or revenue results are available yet.&quot; |\n| &quot;reduced support time to six hours&quot; | Unsupported \u2014 figure is a combined median, not attributed to Pilot B or shown as a reduction | S3 | &quot;The median support response time was six hours across the two pilots combined; it was not broken out by pilot.&quot; |\n\n## Three missing measurements needed for a fair comparison\n\n1. Activation rates for both pilots using an identical definition (e.g., workspace creation *and* at least one teammate invite).\n2. Support response time reported separately for Pilot A and Pilot B (not a combined median), ideally with a pre-pilot baseline.\n3. 30-day retention (and revenue) results for both pilots, collected from comparable, ideally randomized or matched, account populations.<\/code><\/pre><\/details><details><summary>Structured instruction \u2014 complete prompt and source material<\/summary><pre><code>Answer the assignment using only this source packet. Compare definitions, populations, and denominators before treating the two rates as evidence of improvement. Distinguish a draft assertion from measurement. Attach each conclusion to an exact source quote that actually supports it, and leave absent evidence absent.\n\n&lt;task_material&gt;\n# Fictional research packet\n\nAll organizations, data, quotations, and documents below are invented for a reproducible test. Source IDs refer only to this packet; they are not public studies.\n\n## S1 \u2014 Pilot A closeout, 2 September 2026\n\n&quot;Pilot A invited 50 accounts; 40 completed activation within seven days.&quot;\n&quot;Activation meant creating a workspace and inviting at least one teammate.&quot;\n&quot;Accounts were existing customers selected by the customer success team.&quot;\n&quot;The pilot used guided onboarding.&quot;\n\n## S2 \u2014 Pilot B dashboard note, 3 September 2026\n\n&quot;Pilot B recorded 60 onboarding starts; 54 accounts created a workspace within seven days.&quot;\n&quot;This dashboard calls workspace creation activation; it does not require a teammate invitation.&quot;\n&quot;The invitation count was not captured.&quot;\n&quot;Accounts were new self-serve signups using a checklist.&quot;\n\n## S3 \u2014 Analyst review, 4 September 2026\n\n&quot;The pilots were not randomized and used different account populations.&quot;\n&quot;The median support response time was six hours across the two pilots combined; it was not broken out by pilot.&quot;\n&quot;No 30-day retention or revenue results are available yet.&quot;\n\n## S4 \u2014 Draft rollout memo, 5 September 2026\n\n&quot;Pilot B raised activation by 10 percentage points, improved retention, and reduced support time to six hours. Roll out the checklist to every account.&quot;\nThis is a proposal draft, not an additional data source. Its assertions must be checked against S1\u2013S3.\n\n## Assignment shared by both prompt versions\n\nAnswer: Does this packet justify rolling out Pilot B to all accounts?\nUse only S1\u2013S4. Output (1) a recommendation of at most 120 words, (2) a table with columns Claim, Verdict, Source ID, Exact supporting quote, and (3) three missing measurements needed for a fair comparison.\nAudit all three empirical claims in the rollout memo: the activation increase, improved retention, and reduced support time. Use verdicts Supported, Partly supported, Unsupported, or Contradicted, with a short explanation if needed. Include the numerical rates and their denominators in the recommendation or table. Do not invent citations, URLs, causal effects, or findings absent from the packet.\n&lt;\/task_material&gt;\n<\/code><\/pre><\/details><details><summary>Structured instruction \u2014 complete original answer<\/summary><pre><code>## Recommendation (\u2264120 words)\n\nNo \u2014 the packet does not justify a full rollout of Pilot B. Pilot A: 40\/50 accounts (80%) reached &quot;activation&quot; defined as workspace creation *plus* a teammate invite, among existing customers. Pilot B: 54\/60 accounts (90%) created a workspace within seven days, with no invite requirement, among new self-serve signups. The raw 10-point gap (90%\u221280%) mixes different definitions, different populations, and no randomization (S3), so it cannot be read as the checklist &quot;raising&quot; activation. No 30-day retention or revenue data exist yet (S3), so &quot;improved retention&quot; has no evidence. The six-hour support figure is a *combined* median for both pilots, not broken out by pilot, so &quot;reduced support time&quot; has no baseline or pilot-specific support.\n\n## Table\n\n| Claim | Verdict | Source ID | Exact supporting quote |\n|---|---|---|---|\n| &quot;Pilot B raised activation by 10 percentage points&quot; | Partly supported \u2014 the arithmetic gap exists (54\/60=90% vs 40\/50=80%), but the comparison is confounded and cannot support a causal &quot;raised&quot; claim | S1, S2, S3 | S1: &quot;Pilot A invited 50 accounts; 40 completed activation within seven days.&quot; \/ S2: &quot;Pilot B recorded 60 onboarding starts; 54 accounts created a workspace within seven days.&quot; \/ S3: &quot;The pilots were not randomized and used different account populations.&quot; |\n| Activation was measured the same way in both pilots | Contradicted | S1, S2 | S1: &quot;Activation meant creating a workspace and inviting at least one teammate.&quot; \/ S2: &quot;This dashboard calls workspace creation activation; it does not require a teammate invitation.&quot; |\n| &quot;improved retention&quot; | Unsupported | S3 | &quot;No 30-day retention or revenue results are available yet.&quot; |\n| &quot;reduced support time to six hours&quot; | Unsupported | S3 | &quot;The median support response time was six hours across the two pilots combined; it was not broken out by pilot.&quot; |\n\n## Three missing measurements needed for a fair comparison\n\n1. Pilot-specific (not pooled) support response times, before and after the checklist, so any change attributable to Pilot B can be isolated.\n2. 30-day retention (and revenue) figures for both pilots, since none currently exist (&quot;No 30-day retention or revenue results are available yet,&quot; S3).\n3. A single, shared activation definition (e.g., workspace creation *and* teammate invite) applied to comparable, ideally randomized, account populations in both pilots, rather than mixing existing customers (S1) with new self-serve signups (S2).<\/code><\/pre><\/details><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">When a research task becomes a draft, keep the source packet attached to the writing brief. <a href=\"https:\/\/www.glbgpt.com\/home?inviter=hub_content_home&amp;login=1\">GlobalGPT\uc758 \ub2e4\uc911 \ubaa8\ub378 \uc791\uc5c5 \uacf5\uac04<\/a> lets you move between research, writing, and coding in one dashboard. Carry the evidence forward as well as the summary, so the next step has the details it needs.<\/p>\n\n\n\n<div class=\"wp-block-group is-layout-constrained wp-block-group-is-layout-constrained\">\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" width=\"1024\" height=\"645\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-1024x645.png\" alt=\"globalgpt \ub300\uc2dc\ubcf4\ub4dc\" class=\"wp-image-19280\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-1024x645.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-300x189.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-18x11.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-767x483.png 767w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-1536x967.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/glb-2048x1289.png 2048w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<div class=\"wp-block-buttons is-content-justification-center is-layout-flex wp-container-core-buttons-is-layout-3e41869c wp-block-buttons-is-layout-flex\">\n<div class=\"wp-block-button\"><a class=\"wp-block-button__link has-black-color has-text-color has-background has-link-color wp-element-button\" href=\"https:\/\/www.glbgpt.com\/home?inviter=hub_popup&amp;login=1\" style=\"background:linear-gradient(135deg,rgba(250,183,0,0.55) 0%,rgba(255,106,0,0.66) 96%)\"><strong>GlobalGPT\uc5d0\uc11c 100\uac1c \uc774\uc0c1\uc758 \uc778\uae30 \ubaa8\ub378\uc744 \uccb4\ud5d8\ud574 \ubcf4\uc138\uc694<\/strong><\/a><\/div>\n<\/div>\n<\/div>\n\n\n\n<h2 id=\"coding-prompts\" class=\"wp-block-heading\">Claude prompts for coding<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A useful coding prompt describes the behavior, the environment, and what counts as fixed. In a chat, paste the relevant code and error. In Claude Code, reference accessible files and give a concrete verification target. The same text does not turn a chat into a terminal.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/code.claude.com\/docs\/en\/best-practices\">Anthropic\u2019s Claude Code guidance<\/a> recommends giving the assistant checks it can run, such as tests or a build. Its <a href=\"https:\/\/code.claude.com\/docs\/en\/prompt-library\">official prompt library<\/a> treats prompts as starting points to adapt, with an outcome, reference material, and a way to verify the work.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"1105\" height=\"543\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/09\/claude-code-verification_78b457c5049844aa85e144d70072faec.webp\" alt=\"Claude Code documentation recommends giving the model a check it can run.\" class=\"wp-image-19368\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/09\/claude-code-verification_78b457c5049844aa85e144d70072faec.webp 1105w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/09\/claude-code-verification_78b457c5049844aa85e144d70072faec-300x147.webp 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/09\/claude-code-verification_78b457c5049844aa85e144d70072faec-1024x503.webp 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/09\/claude-code-verification_78b457c5049844aa85e144d70072faec-768x377.webp 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/09\/claude-code-verification_78b457c5049844aa85e144d70072faec-18x9.webp 18w\" sizes=\"(max-width: 1105px) 100vw, 1105px\" \/><figcaption class=\"wp-element-caption\">Anthropic recommends concrete checks such as tests, builds, and visual comparisons.<\/figcaption><\/figure>\n\n\n\n<h3 id=\"debug-from-spec\" class=\"wp-block-heading\">9. Debug against explicit behavior and edge cases<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Fixing a small function whose behavior you can define. <strong>Bring:<\/strong> Code, expected behavior, error or failing case, runtime, and constraints.<\/p>\n\n\n\n<style>.bcp-prompt-c1{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-c1 *{box-sizing:border-box}.bcp-prompt-c1 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-c1 p,.bcp-prompt-c1 li,.bcp-prompt-c1 summary,.bcp-prompt-c1 td,.bcp-prompt-c1 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-c1 p{margin:10px 0}.bcp-prompt-c1 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-c1 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-c1 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-c1 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-c1 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-c1 summary{cursor:pointer;font-weight:650}.bcp-prompt-c1 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-c1 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-c1 th,.bcp-prompt-c1 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-c1 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-c1 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-c1 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-c1{padding:16px;margin:18px 0}.bcp-prompt-c1 .bcp-title{font-size:19px}.bcp-prompt-c1 pre{padding:12px;font-size:13px}.bcp-prompt-c1 th,.bcp-prompt-c1 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-c1\"><p class=\"bcp-kicker\">C1 \u00b7 Representative task tested below<\/p><pre><code>Fix the code according to the complete specification. Check validation before early returns, inclusive or exclusive boundaries, and rules that must still apply when an exception is triggered. Preserve [PUBLIC INTERFACE]. Change only what the specification requires.\n\n&lt;specification&gt;\n[VALID INPUTS, INVALID INPUT BEHAVIOR, BUSINESS RULES, AND EXPECTED OUTPUTS]\n&lt;\/specification&gt;\n&lt;code&gt;\n[PASTE CODE]\n&lt;\/code&gt;\n&lt;environment&gt;\n[RUNTIME, DEPENDENCIES, AND WHETHER EXECUTION TOOLS ARE AVAILABLE]\n&lt;\/environment&gt;\n\nReturn the corrected code and a short explanation. Report only verification you actually performed. If execution is unavailable, say so and give the checks I should run.<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Run independent cases. A convincing explanation is not a passing program.<\/p><\/div>\n\n\n\n<h3 id=\"tests-from-spec\" class=\"wp-block-heading\">10. Write tests from a specification<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Finding missing cases before trusting an implementation. <strong>Bring:<\/strong> Behavioral requirements and the project test framework.<\/p>\n\n\n\n<style>.bcp-prompt-c2{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-c2 *{box-sizing:border-box}.bcp-prompt-c2 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-c2 p,.bcp-prompt-c2 li,.bcp-prompt-c2 summary,.bcp-prompt-c2 td,.bcp-prompt-c2 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-c2 p{margin:10px 0}.bcp-prompt-c2 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-c2 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-c2 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-c2 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-c2 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-c2 summary{cursor:pointer;font-weight:650}.bcp-prompt-c2 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-c2 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-c2 th,.bcp-prompt-c2 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-c2 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-c2 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-c2 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-c2{padding:16px;margin:18px 0}.bcp-prompt-c2 .bcp-title{font-size:19px}.bcp-prompt-c2 pre{padding:12px;font-size:13px}.bcp-prompt-c2 th,.bcp-prompt-c2 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-c2\"><p class=\"bcp-kicker\">C2 \u00b7 Adaptable template<\/p><pre><code>Write tests for [FUNCTION OR COMPONENT] using [TEST FRAMEWORK]. Treat the specification below as the contract. Cover a normal case, each stated boundary, invalid inputs, and interactions between rules where they matter.\n\nFor each test, identify the requirement it checks and its expected result. Keep existing requirements intact. If the specification is ambiguous, identify the blocking question instead of guessing. With execution tools, run the relevant tests and report the actual command and result; otherwise label them unrun.\n\n&lt;specification&gt;\n[PASTE SPECIFICATION]\n&lt;\/specification&gt;\n&lt;implementation&gt;\n[PASTE CODE OR REFERENCE AN ACCESSIBLE FILE]\n&lt;\/implementation&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Tests should enforce the specification, not repeat a bug in the current implementation.<\/p><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Keep tests independent of the implementation where possible. For a shipping calculator, test the exact weight thresholds and an invalid input that would otherwise trigger free shipping. Happy-path examples alone miss interactions between rules.<\/p>\n\n\n\n<h3 id=\"narrow-refactor\" class=\"wp-block-heading\">11. Refactor without expanding the job<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Improving a small piece of code while preserving behavior. <strong>Bring:<\/strong> The code, the specific maintenance problem, and compatibility requirements.<\/p>\n\n\n\n<style>.bcp-prompt-c3{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-c3 *{box-sizing:border-box}.bcp-prompt-c3 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-c3 p,.bcp-prompt-c3 li,.bcp-prompt-c3 summary,.bcp-prompt-c3 td,.bcp-prompt-c3 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-c3 p{margin:10px 0}.bcp-prompt-c3 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-c3 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-c3 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-c3 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-c3 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-c3 summary{cursor:pointer;font-weight:650}.bcp-prompt-c3 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-c3 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-c3 th,.bcp-prompt-c3 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-c3 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-c3 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-c3 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-c3{padding:16px;margin:18px 0}.bcp-prompt-c3 .bcp-title{font-size:19px}.bcp-prompt-c3 pre{padding:12px;font-size:13px}.bcp-prompt-c3 th,.bcp-prompt-c3 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-c3\"><p class=\"bcp-kicker\">C3 \u00b7 Adaptable template<\/p><pre><code>Refactor [CODE OR ACCESSIBLE FILE] to improve [ONE SPECIFIC MAINTENANCE PROBLEM]. Preserve [PUBLIC API, RETURN TYPES, ERROR BEHAVIOR, AND OTHER CONTRACTS]. Use existing dependencies only. Do not add unrelated features or redesign adjacent modules.\n\nReturn the smallest useful change, explain any behavior you could not preserve, and identify the tests needed to verify compatibility. If tools are available, run those checks and show the result; otherwise provide the proposed checks and say they have not been run.\n\n[PASTE CODE AND RELEVANT TESTS]<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Review the diff for changes outside the requested scope, then rerun relevant tests.<\/p><\/div>\n\n\n\n<h3 id=\"explain-code\" class=\"wp-block-heading\">12. Explain unfamiliar code through a concrete input<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Understanding a function before modifying it. <strong>Bring:<\/strong> The actual code and one representative input.<\/p>\n\n\n\n<style>.bcp-prompt-c4{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #334155;border-radius:8px;background:#16242a;color:#f3f6f4;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-prompt-c4 *{box-sizing:border-box}.bcp-prompt-c4 .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#fff}.bcp-prompt-c4 p,.bcp-prompt-c4 li,.bcp-prompt-c4 summary,.bcp-prompt-c4 td,.bcp-prompt-c4 th{color:#f3f6f4;font-size:16px;line-height:1.65}.bcp-prompt-c4 p{margin:10px 0}.bcp-prompt-c4 a{color:#b8e9e4;text-decoration:underline}.bcp-prompt-c4 .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#c8dad6;font-weight:700}.bcp-prompt-c4 pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#0e181d;color:#eff6f2;border:1px solid #41535c;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-prompt-c4 code{font:inherit;color:inherit;background:transparent}.bcp-prompt-c4 details{margin-top:14px;border-top:1px solid #52606a;padding-top:12px}.bcp-prompt-c4 summary{cursor:pointer;font-weight:650}.bcp-prompt-c4 .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-prompt-c4 table{width:100%;border-collapse:collapse;font-size:14px}.bcp-prompt-c4 th,.bcp-prompt-c4 td{padding:10px;border-bottom:1px solid #52606a;vertical-align:top;text-align:left;min-width:80px}.bcp-prompt-c4 .bcp-note{font-size:14px;color:#d8e5e1}.bcp-prompt-c4 .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-prompt-c4 .bcp-links a{display:inline-block;padding:8px 12px;background:#294047;border-radius:5px}@media(max-width:600px){.bcp-prompt-c4{padding:16px;margin:18px 0}.bcp-prompt-c4 .bcp-title{font-size:19px}.bcp-prompt-c4 pre{padding:12px;font-size:13px}.bcp-prompt-c4 th,.bcp-prompt-c4 td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-prompt-c4\"><p class=\"bcp-kicker\">C4 \u00b7 Adaptable template<\/p><pre><code>Explain this code to a developer who knows [BACKGROUND] but is new to [LANGUAGE OR PROJECT]. Follow this example input: [INPUT]. Describe the input, relevant branches, state changes, and expected output.\n\nPoint to the actual function names and expressions. Distinguish what the code shows from assumptions about missing dependencies. Finish with one edge case worth testing. Do not rewrite the code or invent behavior for functions whose definitions are absent.\n\n&lt;code&gt;\n[PASTE CODE]\n&lt;\/code&gt;<\/code><\/pre><p class=\"bcp-note\"><strong>Check the result:<\/strong> Verify the described branch and returned value against the code; mark missing dependencies as unknown.<\/p><\/div>\n\n\n\n<style>.bcp-test-c{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-test-c *{box-sizing:border-box}.bcp-test-c .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-test-c p,.bcp-test-c li,.bcp-test-c summary,.bcp-test-c td,.bcp-test-c th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-test-c p{margin:10px 0}.bcp-test-c a{color:#165854;text-decoration:underline}.bcp-test-c .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-test-c pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-test-c code{font:inherit;color:inherit;background:transparent}.bcp-test-c details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-test-c summary{cursor:pointer;font-weight:650}.bcp-test-c .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-test-c table{width:100%;border-collapse:collapse;font-size:14px}.bcp-test-c th,.bcp-test-c td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-test-c .bcp-note{font-size:14px;color:#435853}.bcp-test-c .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-test-c .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-test-c{padding:16px;margin:18px 0}.bcp-test-c .bcp-title{font-size:19px}.bcp-test-c pre{padding:12px;font-size:13px}.bcp-test-c th,.bcp-test-c td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-test-c\"><p class=\"bcp-kicker\">First-output comparison \u00b7 C<\/p><p class=\"bcp-title\">Both returned functions passed all 24 cases.<\/p><div class=\"bcp-scroll\"><table><thead><tr><th scope=\"col\">\ud655\uc778<\/th><th scope=\"col\">Plain instruction<\/th><th scope=\"col\">Structured instruction<\/th><\/tr><\/thead><tbody><tr><td>Independent local cases<\/td><td>24 \/ 24<\/td><td>24 \/ 24<\/td><\/tr><tr><td>Changes within the small function<\/td><td>\uc608<\/td><td>\uc608<\/td><\/tr><tr><td>Claim of tool-based testing<\/td><td>No such claim<\/td><td>Explicitly says no execution<\/td><\/tr><\/tbody><\/table><\/div><p>The plain and structured versions both handled the weight thresholds, free base shipping, remote surcharge, and invalid inputs in our fixed specification. The structured answer explicitly said it had not executed tests. We then ran both functions ourselves; the plain answer had not claimed execution either.<\/p><details><summary>View all 24 inputs and actual pass\/fail results<\/summary><div class=\"bcp-scroll\"><table><thead><tr><th scope=\"col\">[grams, subtotal cents, remote]<\/th><th scope=\"col\">Expected cents or exception<\/th><th scope=\"col\">Plain<\/th><th scope=\"col\">\uad6c\uc870\ud654\ub41c<\/th><\/tr><\/thead><tbody><tr><td>[1, 0, false]<\/td><td>500<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1000, 0, false]<\/td><td>500<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1001, 0, false]<\/td><td>900<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[5000, 0, false]<\/td><td>900<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[5001, 0, false]<\/td><td>1500<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, 9999, false]<\/td><td>500<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, 10000, false]<\/td><td>0<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1000, 10000, true]<\/td><td>300<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[5001, 0, true]<\/td><td>1800<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1001, 9999, true]<\/td><td>1200<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[5001, 10001, false]<\/td><td>0<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[0, 0, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[-1, 0, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1.0, 0, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[&quot;1000&quot;, 0, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[true, 0, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[null, 0, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, -1, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, false, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, 1.5, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, &quot;0&quot;, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, 0, 1]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[1, 0, null]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><tr><td>[0, 10000, false]<\/td><td>ValueError<\/td><td>\ud328\uc2a4<\/td><td>\ud328\uc2a4<\/td><\/tr><\/tbody><\/table><\/div><\/details><p class=\"bcp-note\">Inputs and expected results were fixed before generation. Returned values had to be exact integers; invalid inputs had to raise ValueError. Both functions were inspected and then run in a local process with network access, user-file reads, and file writes denied. Passing these cases is evidence for this function and specification, not a claim of production readiness.<\/p><p class=\"bcp-note\">Elapsed time: 12.58s \/ 13.30s. Reported input + output tokens: 553 + 735 \/ 621 + 1015. One response each; no retries.<\/p><details><summary>Plain instruction \u2014 complete prompt and source material<\/summary><pre><code>Fix the function according to the specification below and provide the requested deliverable.\n\n# Fictional shipping calculator\n\nPython 3. No network or external packages. Fix shipping_cents(weight_grams, subtotal_cents, remote=False) in the code below.\n\nSpecification:\n- weight_grams is a positive integer. subtotal_cents is a nonnegative integer. Python bool values are not valid integers for these two inputs.\n- remote must be a bool.\n- Raise ValueError for any invalid input, even when the order qualifies for free base shipping.\n- Base shipping is 500 cents for 1\u20131000 grams, 900 cents for 1001\u20135000 grams, and 1500 cents above 5000 grams.\n- A subtotal of at least 10000 cents waives base shipping only.\n- Remote delivery adds 300 cents, including orders with free base shipping.\n- Return an int. Preserve the function signature. Do not mutate input data, perform I\/O, or import packages.\n\nBuggy code:\n```python\ndef shipping_cents(weight_grams, subtotal_cents, remote=False):\n    if subtotal_cents &gt;= 10000:\n        return 0\n    if weight_grams &lt; 1000:\n        fee = 500\n    elif weight_grams &lt; 5000:\n        fee = 900\n    else:\n        fee = 1500\n    if remote:\n        fee += 300\n    return fee\n```\n\nDeliverable: one Python fenced block containing the corrected function, then a short explanation of the changes. Do not claim to have run code unless you actually have execution tools. The evaluator will run cases covering the thresholds, free-shipping rule, remote surcharge, and all invalid-input categories in this specification.\n<\/code><\/pre><\/details><details><summary>Plain instruction \u2014 complete original answer<\/summary><pre><code>```python\ndef shipping_cents(weight_grams, subtotal_cents, remote=False):\n    if isinstance(weight_grams, bool) or not isinstance(weight_grams, int):\n        raise ValueError(&quot;weight_grams must be an int&quot;)\n    if weight_grams &lt;= 0:\n        raise ValueError(&quot;weight_grams must be positive&quot;)\n    if isinstance(subtotal_cents, bool) or not isinstance(subtotal_cents, int):\n        raise ValueError(&quot;subtotal_cents must be an int&quot;)\n    if subtotal_cents &lt; 0:\n        raise ValueError(&quot;subtotal_cents must be nonnegative&quot;)\n    if not isinstance(remote, bool):\n        raise ValueError(&quot;remote must be a bool&quot;)\n\n    if weight_grams &lt;= 1000:\n        fee = 500\n    elif weight_grams &lt;= 5000:\n        fee = 900\n    else:\n        fee = 1500\n\n    if subtotal_cents &gt;= 10000:\n        fee = 0\n\n    if remote:\n        fee += 300\n\n    return fee\n```\n\nChanges made:\n1. **Input validation added**: Checks that `weight_grams` and `subtotal_cents` are `int` (excluding `bool`, since `bool` is a subclass of `int`), that `weight_grams` is positive, `subtotal_cents` is nonnegative, and `remote` is a `bool`. Raises `ValueError` otherwise, even if the order would qualify for free shipping.\n2. **Fixed weight thresholds**: Changed `&lt; 1000` and `&lt; 5000` to `&lt;= 1000` and `&lt;= 5000` so the boundaries (1000g, 5000g) fall into the correct tier per the spec (&quot;1\u20131000 grams&quot;, &quot;1001\u20135000 grams&quot;, &quot;above 5000 grams&quot;).\n3. **Fixed free-shipping logic order**: The original code returned `0` immediately when subtotal qualified for free shipping, skipping the remote surcharge entirely. Now the base fee is computed first, zeroed out if the subtotal qualifies, and *then* the remote surcharge is added \u2014 ensuring remote delivery still adds 300 cents even with free base shipping.\n4. **Return type**: All fee values remain plain ints, satisfying the `int` return requirement.<\/code><\/pre><\/details><details><summary>Structured instruction \u2014 complete prompt and source material<\/summary><pre><code>Fix the function according to the complete specification. Apply input validation before any early return, preserve the inclusive thresholds, and keep the base fee separate from the remote surcharge. Provide the requested function and explanation. Report only verification you actually performed.\n\n&lt;task_material&gt;\n# Fictional shipping calculator\n\nPython 3. No network or external packages. Fix shipping_cents(weight_grams, subtotal_cents, remote=False) in the code below.\n\nSpecification:\n- weight_grams is a positive integer. subtotal_cents is a nonnegative integer. Python bool values are not valid integers for these two inputs.\n- remote must be a bool.\n- Raise ValueError for any invalid input, even when the order qualifies for free base shipping.\n- Base shipping is 500 cents for 1\u20131000 grams, 900 cents for 1001\u20135000 grams, and 1500 cents above 5000 grams.\n- A subtotal of at least 10000 cents waives base shipping only.\n- Remote delivery adds 300 cents, including orders with free base shipping.\n- Return an int. Preserve the function signature. Do not mutate input data, perform I\/O, or import packages.\n\nBuggy code:\n```python\ndef shipping_cents(weight_grams, subtotal_cents, remote=False):\n    if subtotal_cents &gt;= 10000:\n        return 0\n    if weight_grams &lt; 1000:\n        fee = 500\n    elif weight_grams &lt; 5000:\n        fee = 900\n    else:\n        fee = 1500\n    if remote:\n        fee += 300\n    return fee\n```\n\nDeliverable: one Python fenced block containing the corrected function, then a short explanation of the changes. Do not claim to have run code unless you actually have execution tools. The evaluator will run cases covering the thresholds, free-shipping rule, remote surcharge, and all invalid-input categories in this specification.\n&lt;\/task_material&gt;\n<\/code><\/pre><\/details><details><summary>Structured instruction \u2014 complete original answer<\/summary><pre><code>```python\ndef shipping_cents(weight_grams, subtotal_cents, remote=False):\n    if isinstance(weight_grams, bool) or not isinstance(weight_grams, int):\n        raise ValueError(&quot;weight_grams must be an int (not bool)&quot;)\n    if weight_grams &lt;= 0:\n        raise ValueError(&quot;weight_grams must be positive&quot;)\n    if isinstance(subtotal_cents, bool) or not isinstance(subtotal_cents, int):\n        raise ValueError(&quot;subtotal_cents must be an int (not bool)&quot;)\n    if subtotal_cents &lt; 0:\n        raise ValueError(&quot;subtotal_cents must be nonnegative&quot;)\n    if not isinstance(remote, bool):\n        raise ValueError(&quot;remote must be a bool&quot;)\n\n    if weight_grams &lt;= 1000:\n        base_fee = 500\n    elif weight_grams &lt;= 5000:\n        base_fee = 900\n    else:\n        base_fee = 1500\n\n    if subtotal_cents &gt;= 10000:\n        base_fee = 0\n\n    surcharge = 300 if remote else 0\n\n    return base_fee + surcharge\n```\n\nExplanation of changes:\n\n1. **Input validation moved before any early return.** The original code returned `0` for `subtotal_cents &gt;= 10000` without validating any inputs first. Now all validation (type checks for `weight_grams`, `subtotal_cents`, and `remote`, including explicit `bool` exclusion for the two integer fields since `bool` is a subclass of `int` in Python) happens before any fee logic or return statement, so invalid inputs always raise `ValueError` even when the order would otherwise qualify for free base shipping.\n\n2. **Inclusive thresholds fixed.** The original used `&lt; 1000` and `&lt; 5000`, which excluded the boundary values (1000 and 5000) from the lower bracket, contrary to the spec&#x27;s &quot;1\u20131000 grams&quot; and &quot;1001\u20135000 grams&quot; inclusive ranges. Changed to `&lt;= 1000` and `&lt;= 5000` so weight exactly at 1000 gets the 500-cent tier and weight exactly at 5000 gets the 900-cent tier.\n\n3. **Base fee kept separate from remote surcharge.** The original computed `fee`, then zeroed it out entirely via early `return 0` when `subtotal_cents &gt;= 10000`, which incorrectly skipped the remote surcharge for orders with free shipping. Now `base_fee` and `surcharge` are computed independently: `base_fee` is set to `0` when the subtotal threshold is met, but `surcharge` (300 cents if `remote`) is always added on top, satisfying &quot;Remote delivery adds 300 cents, including orders with free base shipping.&quot;\n\n4. **Return type.** All fee components are plain ints, and the sum `base_fee + surcharge` is returned as an `int`, matching the required signature and return type.\n\n**Verification actually performed:** None \u2014 I do not have code execution tools available in this environment, so I have not run this code. The above is based solely on manual reading of the specification and tracing the logic by hand; I have not executed test cases against it.<\/code><\/pre><\/details><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">For work that combines code with text, image, or video tasks, you can <a href=\"https:\/\/www.glbgpt.com\/hub\/how-to-use-globalgpt-cli-in-claude-code\/\">connect GlobalGPT CLI to Claude Code<\/a> and bring the platform\u2019s model and media functions into a terminal workflow. Start with an explicit task and the checks you will use to accept the result.<\/p>\n\n\n\n<h2 id=\"prompt-tests\" class=\"wp-block-heading\">What changed when we made the prompts specific?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Extra structure helped one research judgment in this run. It did not remove the added promise from the email or improve the already-correct shipping function. Because both versions received complete requirements, the comparison is a useful reminder: the quality of the brief matters before you start adding more instructions.<\/p>\n\n\n\n<style>.bcp-check-matrix{box-sizing:border-box;max-width:100%;margin:24px 0;padding:24px;border:1px solid #d9ddd8;border-radius:8px;background:#f3f5f0;color:#1d3036;font-family:system-ui,-apple-system,sans-serif;font-size:16px;line-height:1.65;text-align:left}.bcp-check-matrix *{box-sizing:border-box}.bcp-check-matrix .bcp-title{margin:0 0 10px;font-size:21px;font-weight:700;line-height:1.35;color:#172b32}.bcp-check-matrix p,.bcp-check-matrix li,.bcp-check-matrix summary,.bcp-check-matrix td,.bcp-check-matrix th{color:#1d3036;font-size:16px;line-height:1.65}.bcp-check-matrix p{margin:10px 0}.bcp-check-matrix a{color:#165854;text-decoration:underline}.bcp-check-matrix .bcp-kicker{font-size:12px;text-transform:uppercase;letter-spacing:.09em;color:#426059;font-weight:700}.bcp-check-matrix pre{display:block;max-width:100%;overflow-x:auto;white-space:pre-wrap;overflow-wrap:anywhere;word-break:normal;background:#fff;color:#18313a;border:1px solid #d9ddd8;border-radius:6px;padding:18px;font:14px\/1.65 ui-monospace,SFMono-Regular,Consolas,monospace;text-align:left}.bcp-check-matrix code{font:inherit;color:inherit;background:transparent}.bcp-check-matrix details{margin-top:14px;border-top:1px solid #cbd3cd;padding-top:12px}.bcp-check-matrix summary{cursor:pointer;font-weight:650}.bcp-check-matrix .bcp-scroll{max-width:100%;overflow-x:auto}.bcp-check-matrix table{width:100%;border-collapse:collapse;font-size:14px}.bcp-check-matrix th,.bcp-check-matrix td{padding:10px;border-bottom:1px solid #cbd3cd;vertical-align:top;text-align:left;min-width:80px}.bcp-check-matrix .bcp-note{font-size:14px;color:#435853}.bcp-check-matrix .bcp-links{display:flex;gap:10px;flex-wrap:wrap}.bcp-check-matrix .bcp-links a{display:inline-block;padding:8px 12px;background:#e4ebe4;border-radius:5px}@media(max-width:600px){.bcp-check-matrix{padding:16px;margin:18px 0}.bcp-check-matrix .bcp-title{font-size:19px}.bcp-check-matrix pre{padding:12px;font-size:13px}.bcp-check-matrix th,.bcp-check-matrix td{padding:8px;font-size:13px}}<\/style><div class=\"bcp-check-matrix\"><p class=\"bcp-kicker\">Results from the paired examples<\/p><p class=\"bcp-title\">Different tasks need different checks.<\/p><div class=\"bcp-scroll\"><table><thead><tr><th scope=\"col\">Task-specific measure<\/th><th scope=\"col\">Plain instruction<\/th><th scope=\"col\">Structured instruction<\/th><th scope=\"col\">What to do with the result<\/th><\/tr><\/thead><tbody><tr><td>Writing facts and commitments<\/td><td>8 of 9<\/td><td>8 of 9<\/td><td>Remove invented follow-up promises<\/td><\/tr><tr><td>Research evidence checks<\/td><td>6 of 8<\/td><td>7 of 8<\/td><td>Correct the evidence label and compare denominators<\/td><\/tr><tr><td>Python behavior cases<\/td><td>24 of 24<\/td><td>24 of 24<\/td><td>Keep a runnable specification and independent tests<\/td><\/tr><\/tbody><\/table><\/div><p class=\"bcp-note\">These are counts from different checklists, not comparable percentages or an overall model score. The complete prompts, original answers, and test details are under each example. Single-run timing and token counts describe these requests; they do not establish a speed or cost advantage.<\/p><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Use a short prompt when the job and source material are already clear. Add structure when instructions, documents, examples, or competing requirements become difficult to separate. <a href=\"https:\/\/platform.claude.com\/docs\/en\/build-with-claude\/prompt-engineering\/claude-prompting-best-practices\">Anthropic\u2019s guidance also distinguishes model-specific behavior<\/a>, so an old habit of demanding repeated self-checks should not become an automatic addition to every request.<\/p>\n\n\n\n<h2 id=\"improve-prompts\" class=\"wp-block-heading\">How to fix a prompt that gives you a weak answer<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Change the part that caused the problem. If the answer invented a deadline, \u201cmake this better\u201d is too broad. Point to the unsupported detail, supply the boundary, and ask for a targeted correction. Keep the original answer if you are comparing prompt versions rather than silently replacing it.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table><thead><tr><th scope=\"col\">What went wrong<\/th><th scope=\"col\">What to supply or change<\/th><th scope=\"col\">A useful follow-up<\/th><\/tr><\/thead><tbody><tr><td>The answer is generic<\/td><td>A reader, concrete task, and real material<\/td><td>Rewrite for [audience] using only these details: [brief].<\/td><\/tr><tr><td>The voice feels wrong<\/td><td>A sample of your own writing<\/td><td>Match the sentence rhythm and formality of this sample; keep the new brief\u2019s facts.<\/td><\/tr><tr><td>It adds a promise<\/td><td>An explicit boundary and the offending sentence<\/td><td>Remove the follow-up commitment. The source does not authorize one.<\/td><\/tr><tr><td>A citation looks plausible<\/td><td>The actual quoted passage<\/td><td>Does this passage support the full claim, only part, or neither?<\/td><\/tr><tr><td>Two percentages look comparable<\/td><td>Definitions, denominators, populations, period<\/td><td>State what each percentage divides by before comparing the rates.<\/td><\/tr><tr><td>The code looks right but fails<\/td><td>A reproducible failing input and expected result<\/td><td>This input should return [value] but returns [value]. Fix the rule and run this case.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">For a genuinely missing requirement, let Claude ask a focused question. For an ordinary task with enough context, an obligatory twenty-question interview just delays the work. Likewise, ask for a concise explanation, evidence, or test results rather than a transcript of private internal reasoning.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Save useful prompts together with the input shape and the checks that caught mistakes. If the task becomes a repeatable application workflow, <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-api-guide\/\">using Claude through an API<\/a> gives you a place to version prompts and keep request records. Recheck your own examples when the model or task requirements change.<\/p>\n\n\n\n<h2 id=\"faq\" class=\"wp-block-heading\">\uc790\uc8fc \ubb3b\ub294 \uc9c8\ubb38<\/h2>\n\n\n\n<h3 id=\"faq-1\" class=\"wp-block-heading\">What should a good Claude prompt include?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Include the task, relevant context or source material, important constraints, the desired output, and a check for success. For a simple request, a sentence can cover several of these. Use more structure when you are combining documents, examples, or rules that would otherwise be easy to confuse.<\/p>\n\n\n\n<h3 id=\"faq-2\" class=\"wp-block-heading\">Do Claude prompts need XML tags?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No. Clear ordinary language is a useful starting point. XML tags can help separate instructions, examples, and source text in a complex prompt. Use descriptive labels that make the parts easier to follow. Tags do not guarantee factual accuracy, and their presence alone does not establish that one prompt is better.<\/p>\n\n\n\n<h3 id=\"faq-3\" class=\"wp-block-heading\">Can Claude check sources it cannot open?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">It can assess text you provide, but a source title or URL is not enough to establish what a page says. Paste the relevant passage or use available browsing tools to retrieve it. Then check both the quotation and whether that quotation supports the particular claim attached to it.<\/p>\n\n\n\n<h3 id=\"faq-4\" class=\"wp-block-heading\">Can I use the same prompt in Claude chat and Claude Code?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">You can reuse the task and constraints, but adapt the request to the available tools. In chat, provide the code and relevant context. In Claude Code, you can reference accessible project files and request execution. A prompt that says \u201crun the tests\u201d is not evidence that tests actually ran.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The best Claude prompts are the ones you can adapt and verify. Start with one recurring task, replace the sample material with your own, and use the checks beside the template before trusting the answer. <a href=\"https:\/\/www.glbgpt.com\/home?inviter=hub_content_home&amp;login=1\">Try the workflow in GlobalGPT<\/a> when your next step needs another model or AI tool in the same workspace.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Sources and test date:<\/strong> Official prompting and Claude Code documentation were checked on September 14, 2026. The paired examples were generated with Claude in GlobalGPT on that date using fictional material. Explore the complete prompts and answers above, then adapt a template to your own task.<\/p>\n\n\n\n<script type=\"application\/ld+json\">{\n    \"@context\": \"https:\\\/\\\/schema.org\",\n    \"@type\": \"FAQPage\",\n    \"mainEntity\": [\n        {\n            \"@type\": \"Question\",\n            \"name\": \"What should a good Claude prompt include?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Include the task, relevant context or source material, important constraints, the desired output, and a check for success. For a simple request, a sentence can cover several of these. Use more structure when you are combining documents, examples, or rules that would otherwise be easy to confuse.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Do Claude prompts need XML tags?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"No. Clear ordinary language is a useful starting point. XML tags can help separate instructions, examples, and source text in a complex prompt. Use descriptive labels that make the parts easier to follow. Tags do not guarantee factual accuracy, and their presence alone does not establish that one prompt is better.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Can Claude check sources it cannot open?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"It can assess text you provide, but a source title or URL is not enough to establish what a page says. Paste the relevant passage or use available browsing tools to retrieve it. Then check both the quotation and whether that quotation supports the particular claim attached to it.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Can I use the same prompt in Claude chat and Claude Code?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"You can reuse the task and constraints, but adapt the request to the available tools. In chat, provide the code and relevant context. In Claude Code, you can reference accessible project files and request execution. A prompt that says \\u201crun the tests\\u201d is not evidence that tests actually ran.\"\n            }\n        }\n    ]\n}<\/script>","protected":false},"excerpt":{"rendered":"<p>The best Claude prompts make the job easy to understand: give Claude a goal, the material it should use, the constraints that matter, and a way to check the result. For writing, protect the facts and the reader\u2019s next action. For research, connect claims to sources. For coding, define correct behavior and run tests. Start [&hellip;]<\/p>","protected":false},"author":7,"featured_media":19365,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_seopress_robots_primary_cat":"","_seopress_titles_title":"Best Claude Prompts for Writing, Research & Coding","_seopress_titles_desc":"Find the best Claude prompts for writing, research, and coding. Copy 12 adaptable templates, supply your own context, and check the results before you use them.","_seopress_robots_index":"","footnotes":""},"categories":[7],"tags":[],"class_list":["post-19364","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-chat"],"acf":[],"_links":{"self":[{"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/posts\/19364","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/comments?post=19364"}],"version-history":[{"count":3,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/posts\/19364\/revisions"}],"predecessor-version":[{"id":19369,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/posts\/19364\/revisions\/19369"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/media\/19365"}],"wp:attachment":[{"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/media?parent=19364"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/categories?post=19364"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/kr\/wp-json\/wp\/v2\/tags?post=19364"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}