No matter your stance on AI, how often you use it or what you use it for, and how many AI subscriptions are making your bank balance sigh at the end of every month, the word "prompt" has surely made it to your everyday vocabulary by now. Despite that, most of us are still typing some form of "do the thing, but make it good" into a chat box and then wondering why the result feels half-baked.

While you'll find dedicated prompt engineering courses and thousands of posts online promising to turn you into a prompt wizard overnight, I think the best tips come from the people who actually build these models and know exactly how they're meant to be talked to. So I dug into the prompts and prompting principles Anthropic's own engineers swear by, and a few of their habits immediately showed me how much time I'd been burning on bad prompts that were never going to work.

Make Claude Code explain itself in plain English

Built for non-coders, but honestly a lifesaver for the rest of us too

A good chunk of the prompts on this list come from Thariq, who is an engineer on the Claude Code team. He frequently shares tips about his own Claude Code workflow on X, and they tend to be the kind of thing you'd never think to try yourself (but immediately steal once you see them). He also shares prompts from the wider Anthropic team and the community, and this one has a nice origin story.

Thariq shared it after helping his sister vibe-code a WNBA stats site, where he realized that something like 99% of what Claude Code normally says or shows on screen was useless to her since it's all written for people who already speak fluent developer. So, he built a prompt template to fix that, and it works well enough that it's become my default whenever I'm doing something in Claude Code that's outside my depth.

What it does is reshape how Claude Code works and talks. Instead of running everything in your main session and narrating it in jargon, it spins the actual work off to an async background agent — a separate agent that churns away in the background while your main session just checks in on it and reports back.

Alongside that, it puts the explore agent to work: a read-only agent whose only job is to poke around and explain how things function (without touching anything), so you're actually learning what's happening rather than just watching it happen. And every bit of it gets translated into plain English, because the prompt hammers the point that technical language is no use to you.

This is my prompt: "prompt here"

Can you spin off an async background agent to do this and then periodically poll it as it does work and summarize what is happening.

I'm not technical at all so please summarize it in a simple way for me. Use the explore agent to summarize how things work (that the other agent is working on) so I can learn while I do this. If the background agent runs into any errors, please stop and tell me but guide me in how I might be able to fix it.

Remember I am non technical, so any technical language at all is not useful to me.

read more

The prompt that makes Claude quiz you until you actually get it

No more nodding along to code you couldn't explain if asked

Up until a few months ago, the hard part of coding with AI used to be getting the thing to write working code. Now it writes the code fine. The hard part, though, is keeping up with what it just did. As someone majoring in Computer Science who both relies on AI extensively for coding and learning, I'm always looking for ways to ensure I'm actually understanding what the model builds for me, and not just nodding along as it does the work and accepting whatever it hands back.

So when I came across this prompt, shared by Thariq from his colleague Suzanne at Anthropic, it immediately stood out to me. Thariq had been asking people on the team how they actually stay on top of what Claude is building, and this was one of his favorites.

Instead of letting Claude do the work and explain it all in one big dump at the end (the part you skim and instantly forget), this turns Claude into a teacher. It goes one concept at a time, makes you restate things in your own words first, and won't move on until you've actually got it. The part I like most is that it quizzes you, and it does this using AskUserQuestion, a built-in Claude Code tool that lets the model pause and put an actual multiple-choice or open-ended question to you.

The other detail worth pointing out is the last line — that /goal bit. It sets a hard stopping condition: the session isn't allowed to end until Claude has verified you've demonstrated you understand everything on its checklist. No bailing early, no "yeah yeah I get it" and moving on. You either prove it or the lesson keeps going!

you are a wise and incredibly effective teacher. your goal is to make sure the human deeply understands the session.

do this incrementally with each step instead of all at once at the end. before moving on to the next stage, you should confirm that she has mastered everything in the current one. this should be high level (e.g. motivation) and low level (e.g. business logic, edge cases).

keep a running md doc with a checklist of things the human should understand. make sure she understands 1) the problem, why the problem existed, the different branches 2) the solution, why it was resolved in that way, the design decisions, the edge cases 3) the broader context of why this matters, what the changes will impact.

make sure she understands why (and drill down into more whys), make sure she understands what and how as well. understanding the problem well is imperative.

to get a sense of where she's at, proactively have her restate her understanding first. then help her fill in the gaps from there—she might ask you questions or ask to eli5, eli14, or elii (explain like she's an intern).

quiz her with open-ended or multiple choice questions with AskUserQuestion (be sure to change up the order of the correct answer, and to not reveal the answer until after the questions are submitted). show her code or have her use the debugger if necessary!

/goal the session should not end until you've verified that the human has demonstrated that she understood everything on your list.

read more

Make Claude interview you before it builds anything

Turns out the best spec is the one Claude drags out of you

Another prompt from Thariq flips the usual dynamic: instead of you trying to explain everything you want up front (and inevitably forgetting half of it), you make Claude interview you.

The key bit is the AskUserQuestionTool. This is the same interactive-question mechanism from the teaching prompt earlier, except here it's pointed at planning rather than quizzing. Claude reads your spec file, then grills you about the things you didn't think to specify: edge cases, tradeoffs, UI decisions, the stuff that usually only surfaces halfway through building when it's annoying to change. Only once it's done interrogating you does it write the finished spec to the file.

read this @SPEC.md and interview me in detail using the AskUserQuestionTool about literally anything: technical implementation, UI & UX, concerns, tradeoffs, etc. but make sure the questions are not obvious

be very in-depth and continue interviewing me continually until it's complete, then write the spec to the file

read more

If this sounds a lot like Claude Code's Plan mode, that's because it's chasing the same goal from a different angle: get the thinking done before any code gets written, so Claude isn't guessing at your intent. Plan mode has Claude lay out its approach for you to approve; this has Claude pull the approach out of you first.

Make Claude turn your build plan into a real webpage

HTML > Markdown

While this might sound unconventional, I've largely ditched traditional presentations for HTML decks. I've also ditched documents for more interactive, browser-based things I can actually share with a link by hosting them. Thariq shares the exact same sentiment, and he has been very vocal about it. He wrote up an excellent article which he shared on both X and the Claude Blog, where he explained that HTML is quietly a far better default than Markdown for most of what we ask AI tools to produce.

His reasoning is that Markdown is fine for short stuff, but the moment a document runs past a hundred lines or so, nobody's actually reading it. HTML, on the other hand, lets Claude embed real diagrams, tables, syntax-highlighted code, color, and even interactive elements like sliders and copy buttons, and you can drop the file in any browser and share it with someone without them needing a Markdown viewer. He's since said he uses it for planning, speccing, exploration, code review, reports, and a lot more.

This prompt is that philosophy aimed at one specific job: planning out a build before you write any of it. Instead of a wall of text you'll skim once and forget, Claude lays the whole plan out as an HTML page with mock-ups, data flow, and the code snippets actually worth reviewing.

Create a thorough implementation plan in an HTML file, be sure to make some mockups, show data flow, and add important code snippets I might want to review. Make it easy to read and digest.

read more

Turn a recurring task into a skill

Explain it once, trigger it forever

If there's a sequence of steps you find yourself walking Claude through over and over, that's a sign it should be a Skill. I've written a whole piece on why Skills are the most slept-on feature in Claude, so I won't go into the details, but the short version is that a Skill is a set of instructions stored as a portable file, which Claude pulls up automatically when a task calls for it.

This one, from Anthropic's own prompt library, turns a routine you keep repeating into a single command:

create a [/ship] skill for this project that [runs the linter and tests, then drafts a commit message]

read more

The first bracket is the shortcut you'll trigger it with and the the second is the sequence of steps you're tired of walking Claude through every time. It writes the whole thing up as a Skill and it's one anyone on your team can run, too.

Capture what to remember for next time

So Claude stops relearning your project from scratch every session

Every Claude Code session, Claude works out a bunch of small things like how your project is wired, a convention you prefer, a mistake it made and corrected. And then, by default, all of that evaporates when the session ends, so the next one starts from scratch and relearns the same lessons.

This last prompt, also from Anthropic's prompt library, is how you stop that from happening:

summarize what we did this session and suggest what to add to CLAUDE.md

read more

CLAUDE.md is a file Claude Code reads at the start of every session, and it essentially serves as standing memory for a project. With this simple prompt, you don't have to be the one keeping track of what's worth saving. Claude does it for you. It already knows what it had to figure out over the course of the session, so you just let it propose the entries, then keep the ones that actually matter.

The prompts worth actually stealing

While there are a lot of prompts you'll find in Claude's prompt library and floating around online, the ones above are the ones I keep actually reaching for.