25 Prompt Templates
The editor’s picks for September, pulled from Anthropic’s own public prompt library and rebuilt in-house for the jobs our readers actually ask about — writing, coding, business communication, and decision-making. No vague advice, just prompts you can paste in as-is.
Published September 17, 2026
·
18 min read
·
bestprompt.art editorial team
How this list was built. We don’t yet have a live reader-voting feature on bestprompt.art, so despite this month’s working title, nothing here is “reader-voted” — that’s on our roadmap, and we’ll say so plainly again if it launches. What you’re getting instead: 8 templates adapted from Anthropic’s official, public Claude Code prompt library, credited individually below, and 17 templates written in-house by our editorial team, following the role–context–task–format–constraints structure that shows up across most serious prompt-engineering guides. We tell you which is which under each card.
If you’ve read our other tool roundups, you know we’re skeptical of “500 prompts that will change your life” listicles. Most of them are one template copy-pasted 500 times with the nouns swapped. This is the opposite bet: 25 templates, each doing something structurally different, each with a one-line explanation of why it works — so that once you understand the pattern, you can build your own instead of coming back here every month. The five-part shape almost every good template shares
Before the list: nearly every template below (and most templates worth using anywhere) is a variation on the same five slots. Naming them makes it much easier to fix a template that isn’t working, instead of just rewriting the whole thing from scratch.
Role
who’s answering
Context
what it needs to know
Task
the one clear ask
Format
shape of the answer
Constraints
length, tone, limits
Drop any one slot and the model guesses on your behalf — usually wrong.
Every template card below tells you which slots it’s leaning on hardest, so when you adapt one for your own use case, you know which part to rewrite and which part to leave alone.
Where the 25 came from
We’re not going to pretend these were crowd-tested against thousands of submissions — that’s not honest, and it’s not what happened. Here’s the actual split:
Adapted from Anthropic’s public library
8
Written in-house by our editors
17
Bar length is scaled to template count (25 total). Source attribution is repeated under each individual card below.
Section 1 — Ship code faster
Adapted from Anthropic’s public Claude Code prompt library
These eight are lightly edited versions of prompts Anthropic itself publishes for Claude Code users. We didn’t invent the underlying pattern — we picked the eight we reach for most often and added the “why it works” note in our own words.
01
Get oriented in an unfamiliar codebase
Anthropic
Uses Task + Format. You’re not naming files — you’re describing the outcome and letting the model find its own way in.
give me an overview of this codebase: architecture, key directories, and how the pieces connect
Swap in: nothing — this one works as-is on almost any repo.
02
Check what breaks before you delete something
Anthropic
A guardrail prompt. Asking “what depends on this” before you cut code turns a risky delete into an informed one.
what would break if I deleted the retryWithBackoff helper?
Swap in: the function, class, or config key you’re about to remove.
03
Plan a multi-file change before touching code
Anthropic
The phrase “don’t edit anything yet” is doing all the work — it separates exploration from execution, so you can review the plan before any file changes.
plan how to refactor the payment module to support multiple currencies. list the files you would change, but don’t edit anything yet
Swap in: the module and the goal.
04
Write tests, run them, fix what fails
Anthropic
Bundling write + run + fix into one ask means the model iterates on its own instead of stopping after the first draft.
write tests for app/parsers/feed.py, run them, and fix any failures
Swap in: the file or module path.
05
Review your own changes before committing
Anthropic
A pre-commit second pair of eyes that reads full files, not just the diff — it catches the kind of issue a quick self-review misses.
review my uncommitted changes and flag anything that looks risky before I commit
Swap in: nothing needed; add “focus on security” if you want it narrowed.
06
Investigate a production incident
Anthropic
Names the evidence to correlate (logs, deploys, config) rather than prescribing steps — that lets the model cross-reference sources you might not have thought to check.
the checkout endpoint started returning 500s an hour ago. check the logs, recent deploys, and config changes, then tell me the most likely cause
Swap in: the symptom and rough timing. Read More....