Issue #1Received
Copy
Brief 01: the five links we send every new client
First issue. Five pages we end up sending to almost every client in their first week, and what to actually do with each one.
The five links we've sent so many times we keep them in a text file. If you're starting from zero, start here.
1. Anthropic's prompt engineering guide
The vendor's own documentation, which almost nobody reads.
Why: Most of the prompt advice circulating is folklore from 2023 that was never true or stopped being true. This is the primary source, it is short, and the techniques in it are ordered roughly by how much they will help you. If you read one thing on this list, read the section on giving the model a role and examples.
Try: Take the prompt you use most often. Add two worked examples of the output you want, in full, before your instruction. That single change usually does more than an hour of rewording.
2. The Claude Code docs
An agent that works in your terminal, on your files.
Why: The gap between "chatbot that writes code" and "agent that edits your project" is the biggest practical difference in how this stuff gets used, and most people have only seen the first one. This applies well beyond code: any folder of files is a codebase as far as the tool is concerned, which is how we run a lot of document work.
Try: Point it at a directory of messy files you have been avoiding and ask it to describe what is in there before you ask it to change anything. The summary alone is often the thing you actually wanted.
3. Model Context Protocol
How you connect a model to your own tools and data.
Why: This is the answer to "can it read my calendar / my database / my project tracker" without a custom integration per model. Worth understanding even if you never build one, because it tells you what to ask a vendor when they claim their AI integrates with your systems.
Try: Look at the list of existing servers before building anything. The connector you need probably exists, and the ones that do not exist are usually a signal about where the data is genuinely hard to get at.
4. OpenAI's prompt engineering guide
The same subject, from the other vendor.
Why: Reading both is the cheap way to tell technique from house style. What appears in both guides is a real property of these models. What appears in only one is usually specific to that model family, and worth knowing which is which before you carry a habit across.
Try: Skim it right after the Anthropic one and note the three places they disagree. That short list is more useful than either guide alone.
5. Google's own page on replying to reviews
Not AI. The thing the AI is supposed to help you do.
Why: We put a non-AI link on an AI list deliberately. The most common way these projects fail is picking a tool before understanding the task it is meant to do, and review replies are the clearest small example: two pages of rules about what you may and may not say publicly, which no prompt will teach you.
Try: Read it, then go count how many of your reviews have no reply. That number is the project. The tooling is a detail.
Reply if this list was already obvious to you, and we'll aim higher.
Tried one of these? Reply to the email, or write to hey@vetted.tools, and tell us how it went.