Most people use Claude like a search bar: fresh chat, fresh explanation, fresh disappointment. A weekly stack works differently. You pick the six recurring jobs that eat your week, write one strong prompt for each, and then you only ever paste your raw material in. Same input shape, same output shape, no thinking required from you at the start of the task, which is the part that usually makes you avoid it.
The six below are the ones I actually run: a weekly reset, inbox triage, repurposing, a decision memo, meeting notes into actions, and a read on my numbers. Steal the set, then swap any one of them for whatever your week keeps throwing at you.
This is the part that turns six prompts into a system. It takes one sitting and you never do it again.
This is the Sunday one. You paste in your messy notes from the week and next week's calendar, and you get back a clean picture of what actually happened plus a real top three. The thing that makes it work is the fourth section, the list of things to deliberately not do, because that is the decision most weekly reviews skip.
You are my chief of staff and we are doing my weekly reset. Below are my raw notes from this week and my calendar for the week ahead, exactly as they came out, messy and out of order.
Give me four things. First, my wins: what actually got finished, in plain language, no padding and no credit for things that only moved slightly. Second, my open loops: anything I started, promised someone, or am waiting on, with the name of whoever is blocking each one. Third, my top three for next week, chosen by what moves my main goal forward rather than what is loudest, and tell me in one line why each one made the cut. Fourth, what I should deliberately NOT do next week, and what actually happens if I drop it.
If I have not told you my main goal for this quarter, ask me for it before you start. Keep the whole thing under 400 words and write it to me, not about me.
--- MY NOTES FROM THIS WEEK ---
[paste notes, meeting scribbles, anything]
--- MY CALENDAR FOR NEXT WEEK ---
[paste the week view]Paste your unread threads in and get them sorted by what they actually cost you, with a full draft reply for every one that needs it. I do this Monday morning and again Thursday. The rule that matters is the last one: when a thread needs a real decision from you, Claude must say so instead of inventing a position on your behalf.
You are triaging my inbox. Below are my unread email threads pasted in full, including senders and the whole back and forth where I have it.
Sort every thread into four buckets: reply today, reply this week, delegate or forward, and archive with no reply needed. For each thread in the first three buckets give me one line on what the sender actually wants (not what they said), then write a complete draft reply in my voice: warm, direct, short sentences, no corporate filler, no em dashes, no "I hope this finds you well".
Two rules. If a thread needs a decision only I can make, do not guess: say so, lay out the two or three options with the trade-off of each, and give me your recommendation instead of a draft. And put the highest-stakes thread at the top so I start there rather than with the easy ones.
--- MY UNREAD THREADS ---
[paste]One long thing becomes a week of posts, rewritten for each place rather than pasted across all of them. The two details doing the work here are the voice samples and the instruction that each piece must lead with a different idea, which is what stops you saying the same sentence five times in five formats.
Turn the long piece below into a week of content, properly rewritten per platform rather than copy-pasted across them.
Give me: one LinkedIn post (a hook line, short paragraphs, a real opinion, no hashtag spam), three Instagram captions with a different angle each, three short-form video scripts of 30 to 45 seconds where the first spoken line IS the hook, one X thread of five to seven posts, and one short email to my list. Every piece has to stand on its own for someone who never saw the original, and each one must lead with a DIFFERENT idea from the source so I am not repeating myself all week.
Match my voice using the samples below: copy my sentence length, my level of casualness and the words I actually use. At the end, tell me which single idea from the source has the most reach potential and why you think so.
--- THE SOURCE PIECE ---
[paste the long post, newsletter, transcript or article]
--- MY PAST POSTS, FOR VOICE ---
[paste two or three things you wrote]For the choice you have been circling for two weeks. Most AI answers on decisions are mush because nothing forces a conclusion, so this prompt makes Claude argue both sides at full strength, then commit, score its own confidence, and name the one fact that would change its mind.
Write me a decision memo on the choice below. I want both sides argued at full strength and then an actual answer, not a balanced-sounding list that avoids committing.
Structure it like this. One, restate the decision in a single sentence so I know you understood it. Two, the strongest case FOR, including the assumptions it quietly depends on. Three, the strongest case AGAINST, argued just as hard, and include the failure mode I am most likely to be blind to given what I have told you. Four, commit: pick one option, give me a confidence score out of ten, and name the single piece of information that would flip your answer. Five, the smallest reversible step I can take this week to test it before betting properly.
Where you are missing something important, ask me for it rather than filling the gap with an assumption.
--- THE DECISION ---
[what you are choosing between, the money and time at stake, your deadline, what you have already tried, who else it affects]Paste a transcript, get owners, deadlines and the follow-up email. The UNASSIGNED rule is the important bit. A transcript summariser that guesses who owns what quietly creates work nobody agreed to, and you find out three weeks later.
Here is a raw meeting transcript. Turn it into something the room can act on.
Give me a three-line summary of what was actually DECIDED, then a table of action items with the owner, the deadline and the exact wording of what that person committed to. If a name was never said out loud for a task, mark the owner as UNASSIGNED rather than guessing, and list those separately at the end so I can chase them.
Then two more things: the open questions nobody answered, and any decision that got made without the person it affects being in the room. Finish by drafting the follow-up email I will send to everyone. Short, friendly, decisions first, actions in a list, and a closing line asking people to correct anything I got wrong.
--- TRANSCRIPT ---
[paste, or attach the file]The one I run on a Friday. It refuses to just read your metrics back to you, it has to point at the rows that prove the trend, separate a real change from a data problem, and give you one thing to change. Attach the CSV rather than pasting it if the file is big, Claude can read it directly.
You are my analyst. Attached (or pasted below) are my numbers for the last [week / month]. Context: [what the business or account is, and what I am trying to grow].
Do not summarise the numbers back to me. Instead: name the ONE trend that matters most and quote the specific rows or dates that prove it. Then tell me the most likely cause, clearly labelled as a hypothesis rather than a fact. Then give me the single change I should make next, and the exact number I should watch to know whether it worked.
Also flag anything that looks like a data problem rather than real behaviour, like a missing week, a duplicated row, or a metric that jumped in a way humans do not actually behave. Keep the whole thing under 300 words and write it like you are explaining it to me over coffee.
--- MY NUMBERS ---
[paste, or attach the CSV or spreadsheet]Spreading them out is what keeps it sustainable. Nothing here takes more than about ten minutes once the prompt is saved.
A pasted prompt is fine. A skill is better, because Claude reaches for it on its own the moment your request matches, so you stop even opening the note. Ask Claude to build it for you, no code involved.
Turn the prompt below into a reusable Claude skill so I never have to paste it again.
Write me the SKILL.md. Give it a short lowercase-hyphenated name (do not put "claude" or "anthropic" in the name or the upload gets rejected), and a description line that says both what it does AND exactly when to use it, because that line is what makes you trigger it automatically. Then write the instructions as clear numbered steps, and include the rules about tone and format that are in my prompt.
Before you write it, ask me the three or four things you need to know about my format, my inputs and the mistakes you should avoid. When you are done, package it as a zipped folder I can upload.
--- MY PROMPT ---
[paste one of the six]~/.claude/skills/.
What are projects? (Claude Help Center)
What are skills? (Claude Help Center)
Creating custom skills (Claude Help Center)
Chat search and memory (Claude Help Center)