Claude Features
How I Set Up Every Claude Project in 3 Minutes (Copy My Template)
Most people open Claude Projects, click "New Project," type a name, and then sit there staring at the empty Custom Instructions box. Twenty seconds later they close the tab and go back to opening fresh chats and pasting the same context for the fortieth time this month.
I used to do this too. The feature looked powerful, but the blank setup screen was a wall. Without a template, I'd freeze. Without freezing, I'd half-fill the box with a vague paragraph that didn't actually change Claude's behavior. Either way, the Project sat there, mostly unused, while I kept pasting "Act as a senior copywriter…" into new chats.
Then I built a 3-minute setup template. I now use it every single time I create a Project — and I create one for almost every recurring task in my week. The template fills three boxes in a fixed order: name, custom instructions, knowledge base seed. Sixty seconds each. No staring at the blank screen. No half-formed instructions.
Here it is, exactly how I use it.
Why most Claude Projects fail before they're used
Claude Projects is a great feature wearing the disguise of a configuration screen. The interface gives you four boxes (name, description, custom instructions, knowledge base) and asks you to fill them in. There's no template. There's no example. There's no opinion.
So most people do one of three things:
- Skip Custom Instructions entirely, then wonder why their Project doesn't behave any differently from a blank chat
- Paste a 1,200-word system prompt they copied from a Twitter thread, most of which is unrelated to their actual use case
- Write something vague like "You are a helpful marketing assistant" and call it done
None of these unlock what Projects can actually do, which is to compress recurring work into a single click. The value isn't the configuration screen. The value is that every chat in this Project starts with: shared instructions, shared knowledge, shared voice, and zero re-explaining.
If you set the Project up well once, you save fifteen minutes of context-pasting on every chat for the next year. If you set it up poorly, you set yourself up to never use the feature again.
The 3-minute template
Three boxes. Sixty seconds each. Always in this order.
Minute 1 — Name (and one-sentence purpose)
The Project name should be the job, not the topic. "Email Writer" beats "Marketing Stuff" every time. "Reel Hooks for Solo Founders" beats "Content". The narrower the name, the more useful the Project — because you'll know exactly when to open it.
Then in the description field, write one sentence that names who Claude is in this Project and what gets produced. Not three sentences. One.
Senior B2B email copywriter who writes warm, direct outbound emails for SaaS founders selling to operations teams.
That sentence does more work than people realize. It locks in role, audience, output type, and voice in a single line. Claude reads it on every message in the Project, so a vague sentence here means vague output forever.
Minute 2 — Custom instructions (the 4-block template)
This is the one box that everyone gets wrong. It does not need to be 1,200 words. It needs to be four blocks, in a fixed order, each two to four sentences long.
ROLE
[Who Claude is acting as. One sentence. Specific.]
AUDIENCE
[Who the output is for. Their job, their context, their
mental state when they read this. Two sentences.]
VOICE & RULES
[Three voice words OR a paste of two examples that already
sound right. Then 3-5 hard rules: words to avoid, length
limits, structural musts.]
OUTPUT FORMAT
[The exact shape Claude should return. Headers, length,
whether to give one option or three, whether to ask
clarifying questions first.]
Four blocks. Always the same four. The reason this works is that those four blocks cover the full surface area of "what could go wrong in the output" — role drift, audience mismatch, voice mismatch, format mismatch. Cover all four, and the first reply from any chat in this Project is usually 80% of the way there.
Here's a real one I use for an outbound email Project:
ROLE
You are a senior B2B copywriter who has written outbound emails
for 10+ B2B SaaS companies selling to operations teams.
AUDIENCE
The reader is a Head of Operations at a 50-200-person company.
They get 30+ cold emails per week. Their default mental state
when opening any cold email is "skim and delete in 4 seconds."
VOICE & RULES
Voice: direct, dry, lightly conversational. No exclamation
points. No emojis. Match the energy of these two examples:
[paste 2 emails]
Hard rules:
- Never say "circle back," "touch base," "hope this finds you well"
- Subject line max 6 words, lowercase, no punctuation
- Body max 80 words
- One CTA, phrased as a question
OUTPUT FORMAT
Return one subject line + one body. No greeting, no sign-off.
After the email, list 2 alternative subject lines I can A/B.
Do not ask clarifying questions — make confident choices and
note any assumptions in a single line at the bottom.
I wrote that once. Every chat in that Project now ships shippable cold emails on the first try. I edit, I don't rewrite.
Minute 3 — Knowledge base (3-document seed)
The knowledge base is the part most people skip. Don't. Three documents — and three documents is the floor — multiplies the quality of every output in the Project.
The 3-doc seed I add to every Project:
- Voice doc. 4-6 examples of writing that sound right. These can be your own past work, a competitor's site, two great cold emails you got, or three of your favorite tweets. Title it
voice-examples.md. - Don't-do doc. A list of words, phrases, and structural patterns this Project should never use. Title it
do-not-use.md. Mine has 40+ banned phrases. Claude obeys it ruthlessly. - Context doc. Background that Claude needs to actually do the job — your product, your customers, your offers, your pricing, the names of your competitors. Title it
context.md.
That's the floor. From there you add specifics: a swipe file of past hooks for a content Project, a brand book for a brand Project, six anonymized client briefs for a client-work Project. The point is that the Project starts with non-trivial context on day one — not after you've used it for a month.
You can paste these documents directly into the knowledge base, or upload them as files. I prefer pasting because I can edit in place when I notice a banned phrase Claude keeps slipping in.
What 3 minutes of setup actually saves you
Let's run the math on a single Project — the email writer above.
Without it: every email starts with me pasting role + audience + voice rules + format into a new chat. Call it 90 seconds of context paste before I can ask for anything. Then 2-3 rounds of edits because the first output reverts to corporate. Twelve emails per week × 4 minutes of setup-and-edit overhead = 48 minutes a week. About 40 hours a year just on email-prompt overhead.
With it: I open the Project, type one line ("write to a Head of Ops at a fintech who downloaded our compliance checklist last week"), and Claude returns a near-shippable email in fifteen seconds. Edit time drops to one round. Forty hours becomes maybe four.
That's one Project. I have eleven of them. The compounding is the point.
The Pro Agent & Automation Kit ships 15 fully-configured Claude agents — each with the role, voice rules, and knowledge-base structure already done. Drop them in as Projects. Skip the setup minute entirely.
See Pro Agent Kit · $99Five mistakes that quietly ruin Projects
1. Making the Project too broad
"Marketing" is not a Project. "Outbound emails to ops teams" is. The narrower the scope, the higher the output quality, because every block in your Custom Instructions can be sharper. If you find yourself typing four roles into the ROLE block, you're really setting up four Projects.
2. Writing instructions in the abstract
"Be conversational" is meaningless. "Match the voice of these two emails I'm pasting below" is concrete. Examples beat adjectives — every time, in every Project. If your Custom Instructions has zero pasted examples or pasted documents, the Project will produce average output.
3. Skipping the don't-do doc
This is the single highest-leverage document in the knowledge base, and 95% of people don't write it. You're not telling Claude what to do — you're naming the failure modes. "Don't use 'unlock,' 'leverage,' 'seamless,' 'best-in-class.'" Five minutes writing this doc saves an hour of editing every week.
4. Treating the knowledge base as static
Every time you catch Claude making a mistake — using a banned phrase, missing context, getting the audience wrong — go back into the Project and add a line to the relevant doc. The Project compounds in quality only if you maintain it. Two minutes of doc-update per week and the Project is materially smarter by month two.
5. Building one mega-Project for everything
Resist this. A Project that does emails, blog posts, landing pages, and reel hooks will do all four worse than four separate Projects. The whole point is narrow scope and dense context. Mega-Projects dilute both.
When to use a Project — and when not to
Projects are the right tool when:
- You do this work more than once a week
- The job has stable inputs (audience, voice, format don't change)
- You can write down the role, audience, voice, and format rules
- You have at least three reference documents to seed the knowledge base
Projects are the wrong tool when:
- The work is one-off (just open a chat — don't waste a minute on setup)
- The work is heavily exploratory ("help me think through whether to launch X" — Projects narrow you, exploration needs breadth)
- The role would shift mid-conversation (a Project locks you into one role)
For one-offs and exploration, use a regular chat. For everything you do recurringly, build the Project. The 3-minute template makes the second one cheap enough that there's no excuse not to.
One last shortcut
The fastest way to set up a Project is to copy one you've already built. Once you have one Project that's working — say, your email writer — duplicate it, change the four blocks, swap the knowledge base. New Project in under two minutes, because the structure is already there. Within a few weeks you stop building Projects from scratch entirely. You build them by editing.
This is, secretly, the entire workflow. The 3-minute template isn't really about saving 180 seconds. It's about giving you a fixed structure that you'll improve over months — and a starting Project shape that compounds every time you copy it.
Three boxes. Sixty seconds each. Open one. Build one. Then build the next.
Want 15 ready-to-deploy Claude agents — each one drop-in for a Project, with role, voice, and knowledge-base structure already written?
Browse the kits