0 XP
0
G

Ready to practice this for real?

The full reading for this lesson is below, free to anyone. Think of it as the map. A free account adds the guided path: hands-on exercises, saved progress, and each lesson picking up where the last left off. No credit card needed.

Next up: Build Your Weekly Operator

Sign up with email

Already have an account? Sign in

Build Your Weekly Operator

Build a system prompt that turns your recurring Monday deliverable into a 30-second job. Then run it on your own data and keep it.

In From Chatting to Building you met the weekly operator pattern: a system prompt that knows your role, plus fresh data every week, equals the same quality analysis every Monday. This lesson you build yours and run it on your real data before you leave the page.

Two minutes of setup. Open a note and answer three questions:

Your role, in one line. Not your title. "I run a 6-person sales team covering the Southwest" beats "Regional Sales Manager."

Your recurring deliverable. The weekly output with the same shape: pipeline review, content plan, ops update, client status email. Pick the one that eats the most Monday.

Your data source. A CRM export, a dashboard, last week's numbers. Pick something you can get to right now.

Block 1 of your system prompt is role context: everything Claude needs that never changes week to week. Who you are, who reads the output, what decision they make with it.

The Sales Manager: "I manage a 6-person sales team. Every Monday I review our pipeline for my VP, who decides where to spend her deal-support time. Stages: Discovery, Proposal, Negotiation, Closed. A deal is stalled after 21 days in stage."

The Operations Lead: "I run weekly ops review for a fulfillment team. KPIs: order accuracy (target 99.2%), ship time (target 24h), support backlog (under 40). Anything below target 2 weeks straight goes to the COO."

What's in there: definitions, thresholds, the reader, what the reader does next. What's not: adjectives, mission statements.

Block 2 is output format. Describe the exact shape of the deliverable, in order, with lengths. The test: could a stranger produce something that looks like your report from this description alone?

The Sales Manager's block 2: "Output: 1) Three-sentence summary of pipeline health. 2) Deal-by-deal table: name, stage, days in stage, one-line commentary. 3) Top 3 risks this week, each with a recommended action I can take before Friday. Under 400 words. No pep talk, no restating the data back to me."

If you have a past version of the deliverable you were happy with, say so: "Match the structure of the example report below." A real example beats any description of one.

Block 3 is escalation rules. This is what separates an operator from a summarizer. Tell Claude what counts as a problem and what to do about it:

"If any deal over $50K moved backward a stage, flag it at the very top, before the summary."

"If a KPI misses target for the second consecutive week, mark it ESCALATE and draft the one-paragraph COO note."

"If the data is missing a field you need, list it under Open Questions. Never fill gaps with guesses, and never use numbers that aren't in the data I pasted."

That last rule goes in every operator, whatever your role. It's the difference between a report you skim-check and a report you have to audit.

Now assemble and run it. Stack the three blocks in order: role context, output format, escalation rules. End with one line: "Each week I'll paste fresh data. Produce the deliverable. Ask me nothing you can answer from the blocks above."

Run it now: paste the operator as the first message of a new conversation, or into Project instructions so every future conversation inherits it. Then paste this week's real data, redacted.

Three checks on the output: every number traces back to the pasted data; the format matches block 2, section for section; the recommendations are actionable today. "Follow up with Meridian about the stalled proposal" passes. "Monitor at-risk deals" fails.

Any check fails? Fix the block that let it happen, not the output, and run again.

This is yours. Save it now. Put the corrected prompt somewhere you'll find it next Monday: a note titled "Weekly Operator," a doc, or Project instructions (already saved, in that case).

Next Monday the entire job is: open the operator, paste fresh data, read the output against your three checks. Thirty seconds of your input for the deliverable that used to take the morning. When the output drifts, edit the block, not the draft.

And the skeleton transfers: client status email, monthly board summary, sprint recap. The second operator takes minutes because you already know your blocks.

You built a weekly operator: role context, output format, escalation rules, assembled into one system prompt and run against your real data. You know the three checks (numbers trace to the data, format matches, recommendations are actionable) and the repair rule: fix the block, not the output. It's saved where you'll find it. Next Monday it costs you thirty seconds.

Nightschool AI is an independent learning platform and is not affiliated with, endorsed by, or sponsored by Anthropic. Claude is a trademark of Anthropic, PBC. Looking for Anthropic's official Claude Academy? It's at academy.claude.com.

Build Your Weekly Operator: The Practitioner Setup | Nightschool AI