Development

Five Prompts I Reuse Every Week to Ship Code Faster

Kevin PlumleeSeptember 14, 20264 min read
Five Prompts I Reuse Every Week to Ship Code Faster

You do not need a hundred prompt tricks.

Five reusable briefs cover most weekly coding work. They are boring on purpose. Boring is how you ship without the agent rewriting your app because it felt inspired.

These are not magic words. They are constraints that prevent the same failure modes from showing up again.

1) The small-diff bug brief

When something is broken, refuse to say “fix this.”

Paste:

  • what you expected
  • what you saw
  • exact error text
  • the file you already suspect
  • “Propose the smallest diff. Do not refactor neighbors. List assumptions.”

That last line saves you from a 40-file “cleanup” that somehow also renames half your variables.

If the agent wants a rewrite, say no the way you would to a junior on a Friday afternoon. Small patch. Then tests. Then talk about architecture when the building is not on fire.

2) The explainer before the edit

Before you let it touch code, ask for a map.

“Read these files. Explain the current flow in plain English. Call out the risky seams. Do not change anything yet.”

This does two things. It forces the model to show what it thinks is true. And it gives you a chance to correct the story before it starts swinging a hammer.

A lot of bad PRs start with a wrong mental model and very confident typing.

3) The test-first squeeze

Use this when the behavior is clear but the implementation is fuzzy.

“Write failing tests for this behavior. Do not implement yet. Cover happy path, empty input, and one nasty edge case.”

Then review the tests like a human. Then ask for the implementation to make them pass with minimal changes.

Agents love to invent APIs. Tests nail the contract to the floor.

4) The PR self-review

Before you open a pull request, make the agent critique its own work with a checklist.

“Review this diff as a senior engineer. Look for: missing error handling, secrets, silent catches, N+1 queries, unclear names, tests that assert nothing, and scope creep. Output findings only. No praise.”

No praise matters. Models will compliment you for breathing if you let them. You want findings.

Still do your own review. The agent is a second set of eyes that does not get tired of looking at try/except blocks.

5) The “done means” ship brief

When you start a feature, write acceptance criteria like you are briefing a contractor who bills by the hour.

Template:

  • Goal in one sentence
  • Done means (bullets)
  • Out of scope
  • Files or modules to prefer
  • Commands to run
  • Output: plan first, then patch

“Done means” is the whole prompt. Without it, the agent optimizes for looking busy. With it, you can reject work that is impressive and useless.

How to keep them from rotting

Store these as snippets in your notes, not as a sacred prompt library website. When a failure mode shows up twice, add one sentence to the relevant brief.

Examples of sentences that earn their keep:

  • “Do not add new dependencies without asking.”
  • “Match existing logging style.”
  • “If you are unsure, ask. Do not invent env vars.”
  • “Prefer editing the function in place over extracting six helpers.”

Your codebase has a culture. The prompt should carry that culture, or the agent will invent a new one every session.

What not to bother prompting for

Do not spend energy on “act as a world-class 10x ninja.” That does nothing measurable.

Also do not ask for giant greenfield rewrites in one shot. If you need a new module, still slice it: scaffold, core path, tests, then polish.

And never let “make it production ready” stand alone. Production ready according to whom? Say the checks.

A simple weekly rhythm

Monday: explainer prompt on a gnarly webhook handler before you touch it. Tuesday: small-diff bug brief on a timezone bug that only happens on Sundays. Wednesday: test-first squeeze for a new CRM field sync. Thursday: feature ship brief for an admin filter. Friday: PR self-review before you ask a human teammate to look.

Same five shapes. Different files. Fewer surprises.

The human part still matters

These prompts make agents useful. They do not make you optional.

You still decide product tradeoffs. You still reject cleverness. You still run the app with your hands. If you skip that, you are not “AI native.” You are just merging faster into regret.

Bottom line

Copy the five shapes. Inject your stack, your test command, your “never touch” list.

The win is not witty phrasing. The win is repeating a brief that keeps the agent in a lane you trust. Lanes are how you go fast without becoming a cautionary tale in Slack.

Kevin Plumlee

Founder, RVNOPS

Kevin Plumlee

15+ years running healthcare SEO, paid media, and operations. Now I advise the teams who do that work, across marketing, intake, technology, and AI.

More about me

Tell me what’s not working.

Stalled growth, spend you can’t trace, systems that fight your team. That’s a good place to start.

Start a conversation