What you’ll make
Save a short Claude brief you can reuse for one job.
Worked example
A support brief for a downloadable icon pack
A customer asks which files to use if they want to change an icon’s color. The shop owner supplies a file guide that explains where to find the editable SVGs and how to open them in the editor they support.
The reply points the customer to the SVG folder and uses the verified editor instructions. In a second test, the setup note leaves out the supported editor. Claude asks for that detail before giving steps.
Before you start
- Pick a real task that takes you several attempts today, such as answering support questions from product documentation. Bring an anonymized request and an example of an answer you approved.
- Check which account or workspace you are signed into. Start with the features you can see today and make a note if a specific model, file workflow, or tool is essential to your test.
Pick one job to make easier
Think of a request you’ve answered more than once. Write down what you start with, what a usable answer looks like, and the reference you check. Use a practice case without customer details.
I start with a message asking how to change an icon’s color and my file guide.
I want a reply that starts with where to find the SVGs and uses the steps for the editor I support.
Turn those notes into a brief
Copy the prompt into a new Claude chat. Attach the file guide or paste the lines you want it to use, then send it. If Claude asks a question that changes the brief, answer from your source.
I give Claude the color question and the editor instructions in my file guide.
I might get a short brief that says what to attach each time, how to open the reply, and what to ask when the editor is unclear.
Write my reusable Claude brief
Write a short reusable brief for one task I do in Claude. Use the details I provide in this chat or in my attachments. Make it ready to paste into a fresh chat. State the job, what I supply each time, which source to trust for product or policy details, the shape of the answer, and when to ask a question. Preserve the practical checks I gave you. Leave case-specific facts for the individual request. Ask one focused question if you can’t write a usable brief from what I supplied. Return just the brief once the task is clear.
Save the brief where you’ll find it
Read the brief beside your source. Keep the rules that apply to the next request, such as checking the editor first. Save the brief in a note or a Claude Project. Add each new customer question in its own chat.
Try it once in a fresh chat
Paste the saved brief, add your practice question, and supply the reference it names. Read the reply next to the guide and fix the brief if the answer goes beyond the source.
- The reply starts with a useful next action.
- Any product steps match the approved guide.
- The saved brief makes sense with a new question.
Optional: Turn a working brief into a repeatable setup
Once a brief works in a fresh chat, choose where to keep it. Put instructions for this one job in a Project if you’ll use it again, then test a normal request and one with a missing detail.
Set Claude up for one job I actually do
Use this with a sample task to get a short set of preferences and a reusable brief.
I’m setting up Claude for [repeat job]. Here’s a real-looking sample input with private details removed: [sample]. The result I need is [artifact I can use]. It’s good if [two or three things I can check]. My general preferences are [tone, structure, units, or what to do when unsure]. Give me two paste-ready blocks. First, write two or three account-wide preferences that would help in any chat. Then write a reusable brief for this job with required inputs, approved references, the exact output shape, and which details Claude should ask me about before making a claim. Include one short model answer using only my sample. If there’s a gap, mark it and ask me one question rather than filling it in. Give me one ordinary request I can use to see whether the preferences helped.
Test it with a full example and a missing fact
Use this after setup to find out whether the brief holds up in a fresh chat.
Here’s my brief and approved source material: [paste both]. Help me prepare a two-run test for [job]. Run A should have everything Claude needs. Run B should be almost identical, except one fact that changes the answer is missing. Write the two paste-ready requests and a checklist I can use to judge them. For Run A, name the source fact the answer must preserve and the format it must follow. For Run B, show the exact kind of question or placeholder I should see instead of a guess. If an answer fails, tell me which instruction I could change first and what to rerun.
Troubleshooting
Why do unrelated chats sound like support replies?
Move instructions specific to support into the project or reusable brief. Leave only your broadly useful preferences in account settings.
Why did Claude assume a product detail?
Place that detail in the approved reference with a version date. Add a rule that missing policy or product facts require a question, then rerun the incomplete test.