50 Prompts That Do Real Engineering Work

Most prompt lists are written by people who do not ship code.

This one is different. Fifty prompts, five jobs, all tested on real work. Swap the parts in brackets and go.

One tip before you start: give the model your constraints, not just your wish. The prompts below all do this. That is why they work.

Code review (1-10)

1. The second reviewer: "Review this diff like a staff engineer who did not write it. List only issues that would break production or bite us in six months. Skip style. [paste diff]"

2. The security pass: "Review this code only for security: injection, auth bypass, secrets, unsafe deserialization, SSRF. For each finding, show the exact line and a one-line fix. [paste code]"

3. The edge-case hunter: "List the 10 inputs most likely to break this function. Then tell me which ones the current code already handles. [paste function]"

4. The concurrency check: "Assume two requests hit this code at the same time. Walk through what happens step by step. Where does it race?"

5. The API contract check: "Here is the old response shape and the new one. List every way this change can break an existing client. [paste both]"

6. The test gap finder: "Here is the code and here are its tests. What behavior is untested? Rank by how expensive each gap would be in production."

7. The performance pass: "Find the N+1 queries, unnecessary allocations, and anything that gets slower as data grows. Assume this table has 10 million rows."

8. The error-path review: "Trace every failure path in this code. Which errors are swallowed? Which leave the system in a half-done state?"

9. The naming review: "Which names here lie about what the thing actually does? Suggest honest ones."

10. The rollback question: "If we ship this and it goes wrong, how do we roll back? Does anything here make rollback impossible?"

Debugging and incidents (11-20)

11. The rubber duck: "I expect [X]. I observe [Y]. Here is the code and the logs. List the 5 most likely causes ranked by probability, and the fastest check for each. Do not guess. Ask me for missing info first."

12. The log translator: "Here is a stack trace. Explain what actually failed in one sentence, then the most likely cause, then the file I should open first. [paste trace]"

13. The bisect planner: "This worked on [date] and is broken now. Here is the list of merged PRs between. Which 3 are the most likely suspects and why? [paste list]"

14. The heisenbug interview: "This bug happens in production but never locally. Ask me the 10 questions that would best narrow down the difference."

15. The timeout detective: "Requests time out under load but work fine at low traffic. Given this architecture, list where the queue is building up. [describe stack]"

16. The memory leak triage: "Memory grows over hours and drops on restart. Here is the code that changed recently. What holds references longer than it should?"

17. The incident timeline: "Here are raw notes and log lines from an incident. Build a clean timeline: what happened, when, what we knew, and what we did. [paste notes]"

18. The postmortem draft: "Write a blameless postmortem from this timeline: summary, impact, root cause, contributing factors, action items with owners. Keep it under one page."

19. The five whys: "The immediate cause was [X]. Run five whys on it, but stop at causes we can actually fix with process or code, not 'human error.'"

20. The regression guard: "Write the single test that would have caught this bug before deploy. Smallest test possible. [paste bug and fix]"

Architecture and decisions (21-30)

21. The design review: "Here is a design doc. Attack it: what breaks at 10x scale, what is the riskiest assumption, and what would a senior skeptic ask in review? [paste doc]"

22. The build vs buy: "We are deciding whether to build [X] or buy [vendor]. Our team is [size], our real constraint is [time or money or control]. Argue both sides honestly, then pick one."

23. The migration plan: "Plan a migration from [A] to [B] with zero downtime. Give me phases, the rollback point for each phase, and what can run in parallel."

24. The boring alternative: "Here is our exciting new architecture. Propose the boring version that gets 80% of the value with 20% of the moving parts. [paste design]"

25. The failure modes: "List every way this system fails when a dependency goes down. For each: blast radius, detection time, and recovery path. [describe system]"

26. The ADR writer: "Write an architecture decision record for this choice: context, decision, alternatives considered, consequences. One page. [describe decision]"

27. The tech debt case: "Help me argue for paying down [debt item] to a non-technical exec. Translate the cost into missed features, incidents, and hiring pain. Real numbers where I give them."

28. The API designer: "Design the API for [feature]. Optimize for the caller's code being simple, not ours. Show me example calls first, then the schema."

29. The capacity math: "Given [requests/sec, payload size, growth rate], estimate what we need in compute, storage, and bandwidth for 12 months. Show the arithmetic."

30. The kill criteria: "We are adopting [tool]. Define the measurable conditions under which we would rip it out in 6 months. If we cannot define them, say so."

Docs and communication (31-40)

31. The README: "Write a README for this repo: what it does in one sentence, how to run it in under 5 commands, and the 3 things that will confuse a new engineer. [paste tree and main files]"

32. The PR description: "Write the PR description for this diff: what changed, why, how to test it, what could break. No filler. [paste diff]"

33. The exec summary: "Summarize this technical doc for a CEO in 5 sentences: what it is, what it costs, what we get, what happens if we do nothing. [paste doc]"

34. The onboarding doc: "Turn these scattered notes into a day-one onboarding guide. Order it by what a new hire needs first, not by system. [paste notes]"

35. The changelog: "Turn these commit messages into a changelog humans want to read. Group by user impact, not by module. [paste log]"

36. The meeting compressor: "Here is a meeting transcript. Give me decisions made, action items with owners, and open questions. Nothing else. [paste transcript]"

37. The bad news email: "Help me tell stakeholders the launch slips 3 weeks. Lead with the new date, own the cause in one sentence, and end with what we are doing differently. No groveling."

38. The runbook: "Write the 3 a.m. runbook for this alert: what it means, what to check in order, when to escalate, and the one command that usually fixes it. [describe alert]"

39. The interview question: "Design an interview exercise that tests whether a candidate can actually do [job task]. It must fit in 45 minutes and not require trivia."

40. The glossary: "Extract every internal term and acronym from this doc and define each in one sentence a new hire would understand. [paste doc]"

Leadership and leverage (41-50)

41. The delegation test: "Here is my task list for the week. Which items can only I do, and which am I hoarding? Be blunt. [paste list]"

42. The one-on-one prep: "Here are my notes on [person]: recent wins, struggles, goals. Give me 5 questions for our one-on-one that are not 'how is it going.'"

43. The priority fight: "Two teams both claim their project is P0. Here are both cases. Build the comparison a VP would actually use to decide. [paste both]"

44. The estimate check: "The team says [feature] takes 6 weeks. Ask me the 10 questions that reveal whether that estimate is padded, honest, or fantasy."

45. The hiring bar: "Here is a job description. Cut it to the 5 requirements that actually predict success and delete the wish list. [paste JD]"

46. The vendor negotiation: "We are renewing [vendor] at [price]. List the leverage we have, the questions to ask, and the concessions vendors in this space typically give."

47. The skip-level question: "I have 30 minutes with my skip-level. Given my goals [X], what 3 topics get the most value from that half hour?"

48. The strategy translator: "Company strategy says [strategy]. Translate that into what my platform team should start, stop, and continue this quarter."

49. The performance review: "Turn these bullet points about [person]'s work into review feedback that is specific, fair, and includes one growth area with a concrete next step. [paste notes]"

50. The career ladder: "I am a [level] who wants [next level]. Here is what I shipped this year. What is the gap, in the words a promotion committee would use? [paste list]"

How to get more out of every prompt here

Three habits separate people who get real work out of AI from people who get slop:

• Paste real artifacts. Diffs, logs, docs. The model is only as grounded as what you feed it.

• Give constraints. Team size, deadline, scale. Constraints turn generic advice into your answer.

• Ask for ranked lists, not essays. Ranking forces the model to commit.

This library is a Codeinate welcome gift. Two issues a week, Monday and Thursday, five minutes each. If a friend sent you this link, get your own at the homepage.