📝 Prompt Library

Explain [TOPIC] to someone who is intelligent but completely new to the field. Use one concrete analogy, define every term of art the first time you use it, and end with the three most common misconceptions.

Review this code as a senior engineer. Only report real problems: bugs, security issues, race conditions, and misleading names. For each finding give the line, why it matters, and a minimal fix. Do not comment on style. [CODE]

I will describe a bug. Ask me one question at a time — the question most likely to halve the search space — until you can name the root cause. Do not guess early. Bug: [DESCRIPTION]

Summarize this paper at three levels: (1) one tweet, (2) one paragraph for a practitioner, (3) half a page covering method, evidence, and limitations. Flag any claim the abstract makes that the results do not support. [PAPER TEXT OR ABSTRACT]

I believe: [CLAIM]. Argue the strongest possible case against it — steelman, not strawman. Then tell me which single counterargument should worry me most and what evidence would settle it.

Turn these rough notes into a one-page decision doc: context (3 sentences), options considered with one-line trade-offs, recommendation, risks, and what would change the decision. Plain language, no filler. [NOTES]

Write a 90-word cold email to [PERSON/ROLE] about [OFFER]. First line must reference something specific about them. One clear ask. No buzzwords, no “I hope this finds you well”. Give 3 subject-line options under 6 words.

[SUBJECT], shot on 85mm f/1.4, golden-hour rim lighting, shallow depth of field, natural skin texture, color graded like Kodak Portra 400, photorealistic — no text, no watermark, no extra fingers.

Single continuous shot: [SCENE]. Camera: slow dolly-in at eye level. Lighting: motivated practical light from [SOURCE]. Mood: [MOOD]. Duration ~6s. No cuts, no morphing, stable subject identity throughout.

Schema: [SCHEMA] Question: [BUSINESS QUESTION] Write the SQL. Before the query, state your assumptions in one line each. After it, sanity-check: what result size do you expect and what would indicate the query is wrong?

From this transcript extract: decisions made (with who decided), action items as “owner — task — due”, and open questions nobody answered. Mark anything ambiguous with [?] rather than guessing. [TRANSCRIPT]

Help me write a production system prompt for an AI agent that [JOB]. Include: role, hard constraints, output format, refusal rules, and 2 worked examples. Then attack your own prompt: list 3 inputs that would break it and patch them.