Ten quick systems, five weekend agents and five bigger projects you can build in Claude or ChatGPT. Every one has a copy-paste prompt and a clear first step.
19 copy-paste promptsClaude, ChatGPT or Codex10 quick builds + 10 weekend buildsBuilt by Brooke
The five weekend agents from the video: inbox triage, meeting-to-action, second brain, competitor watcher and repurposing engine
Five bigger weekend projects, with a prompt for each build
The original ten systems and their copy-paste prompts
Where connected tools help, and how to start when they are unavailable
The five agents from the video, five bigger projects and ten quicker systems. Pick one and use the prompt to build a first version with your real information.
The ten quick systems start with a prompt. The weekend agents and bigger projects may also need files, connected apps or a place to keep the result. Start with one project and connect only what it actually needs.
1
Open Claude
Web or the desktop app, either works. A free account is enough to try most of these.
2
Turn on your connectors
Check which connections your account offers, then enable only the ones your chosen build needs. The Life Dashboard may need Calendar, email and tasks; the Second Brain may need file access. If a connection is missing, upload a small test set instead.
3
Make a Project
Projects keep your instructions and files in one place, so you're not re-explaining yourself every chat. Make one called My systems and build them all in there.
One thing before you paste anything
These prompts are deliberately bossy. The bluntness is doing real work — lines like "don't be nice about it" and "do not fill gaps with guesses" are what stop you getting a polite, useless answer. Don't soften them.
Five weekend agents
These are the five from my newest video. Each one has a prompt you can paste into Claude or ChatGPT. Build the first version by hand with real information, check it, then put it on a schedule.
1
Inbox triage agent
read → sort into three → draft the replies
Your inbox, sorted before you open it.
It reads your email every morning and splits it into three piles: reply now, reply this week, or ignore. Then it drafts the replies for the first two piles, so you are checking and sending instead of writing from scratch. System #9 below is the four-pile version if you want more detail.
Go through my email from the last 24 hours and sort every message into one of three piles:
REPLY NOW: it needs an answer from me today. One line on why.
REPLY THIS WEEK: it needs an answer, but not today. Suggest which day.
IGNORE: newsletters, receipts, notifications and anything that needs nothing from me.
For every message in REPLY NOW and REPLY THIS WEEK, write the draft reply. My voice: warm, direct and short. Get to the point in the first line. If you can't tell what I'd want to say, write the question you need me to answer instead of guessing.
Save the replies as drafts. Do not send anything. Do not archive or delete anything.
Finish with a count for each pile and the three REPLY NOW messages I should open first.
Run it by hand for a week before you put it on a schedule. Check which messages it put in IGNORE, because that pile is where mistakes hide.
Works in:Claude (connect Gmail, then save it as a morning routine) · ChatGPT (connect Gmail and use a scheduled task) · Both can write drafts without sending anything
2
Meeting-to-action pipeline
recording → decisions + owners + follow-up email
The follow-up is written before you have closed the call.
Drop in the meeting recording or transcript. You get back what was decided, who owns what and a follow-up email ready to send. This is the after-the-call partner to the meeting prep system (#4) below.
Here is my meeting: [ATTACH THE RECORDING OR PASTE THE TRANSCRIPT]
The meeting was with [WHO] about [WHAT].
Give me:
DECISIONS: what was actually agreed. Quote the line that shows it.
ACTIONS: every task, who owns it and the due date. If nobody took it, write "no owner" and flag it. If no date was set, write "no date".
OPEN QUESTIONS: anything raised but not settled.
FOLLOW-UP EMAIL: a short email to everyone in the meeting with the decisions and actions. Write it in my voice and ready to send. Don't add anything that wasn't said.
Only use what is in the recording or transcript. If you can't tell who said something or who owns a task, say so instead of guessing.
Check the owners before you send the email. A task with the wrong name next to it is how things fall through the cracks.
Works in:Claude or ChatGPT with a transcript, which most meeting recorders export. If your plan can read audio or video files, you can upload the recording itself. If it can't, get the transcript first.
3
AI second brain
one library → grows every time you use it
Drop everything in once. Find it again whenever you need it.
This is a library of context you keep adding to: notes, contacts, documents and whatever else you want to find later. The prompt is Second Brain, project 2 in the next section. For the memory of your business itself, pair it with system #10.
Start with five files, check every answer against the original, then keep adding. It gets more useful every week you feed it.
4
Competitor watcher
your list → weekly check → what changed + what's working
Know what is working in your niche without scrolling for it.
You give it a list of accounts to check every week. It reports back what has changed and what is working. Mine runs on a schedule, so I have content ideas ready on Sunday night.
You're my competitor watcher. Every week, check these accounts and pages: [LIST OF ACCOUNTS, WEBSITES OR CHANNELS]
For each one, tell me:
WHAT CHANGED: new posts, offers, prices, launches or messaging since last week. Link every item.
WHAT'S WORKING: the posts getting clearly more engagement than their usual, and your best guess at why.
Only report what you can actually see. If you can't open an account or read its numbers, write "couldn't check" instead of guessing.
Then give me:
THE PATTERN: anything showing up across more than one account.
MY IDEAS: three content ideas for me, based on the gaps or patterns you found. Don't copy anyone.
Keep this week's report so you can compare it with next week's.
Some platforms don't let AI see everything. Test it once and look at what came back as "couldn't check". Only put it on a schedule once the report is worth reading.
Works in:Claude (turn on web search, then set it as a weekly routine) · ChatGPT (search, then a scheduled task) · What it can see depends on each platform, so test it with your real list first
5
Repurposing engine
one long video → short clips + captions in your voice
Film once. Get a week of shorts.
One long video goes in. It finds the best moments to cut into shorts and writes the captions in your voice. System #3 below does the written side of this (posts, emails and a carousel) from the same video.
Here's the transcript of my long video, with timestamps: [PASTE TRANSCRIPT OR ATTACH THE FILE]
Find the 5 best moments to cut into short clips. Each one needs to make sense on its own, with no context from the rest of the video.
For each clip give me:
START AND END TIME: taken from the transcript. Aim for 30 to 60 seconds.
THE HOOK: the first line someone hears, and why it stops the scroll.
ON-SCREEN TEXT: 3 to 6 words.
CAPTION: in my voice. Short, and one clear point. No hashtag wall.
If there aren't 5 strong moments, give me fewer and say so. Don't pad it. Only quote what I actually say.
Here's how I write: [PASTE 2 OR 3 OF YOUR CAPTIONS, OR YOUR VOICE GUIDE FROM SYSTEM #7]
The timestamps are the whole point. Cut the clips yourself (or in your editor) from those times, and check each one makes sense without the rest of the video.
Works in:Claude or ChatGPT with the transcript. Doing the actual cutting automatically needs an editing tool or a coding agent like Codex or Claude Code working on the file.
Five bigger weekend projects
These five bigger projects are from my other weekend video. Each has a prompt you can paste into Claude or ChatGPT. Pick one, make the first version with real information, then test it before you add automation. The ten shorter systems are just below.
1
Life Dashboard
calendar + inbox + tasks + morning brief
One place to see the day before the day starts.
Pull your real calendar, email and tasks into one view. The useful bit is the short briefing at the top: what needs your attention, what can wait and what might clash.
Help me build a one-page Life Dashboard for my real day.
First, tell me which of these you can actually access in this chat: my calendar, email, task list and any connected tools. Ask me to connect or supply only what is missing. Do not invent events, messages or tasks.
Build a first version with these panels:
TODAY — meetings, times and preparation needed
INBOX — messages that need a decision or reply
TASKS — due today, overdue and waiting on someone else
MORNING BRIEF — my top three priorities, any time clashes and the one thing I should prepare before the first meeting
Show a source and last-updated time for every live item. If you can make an interactive page here, build it. If not, give me a clear one-page layout I can paste into a document or dashboard.
After we check the first version against the source data, show me how to refresh it each morning using the features available in my account. Do not turn on a schedule yet.
Start with a real read of one day. A pretty dashboard with made-up data is a picture, not a system.
Works in:Claude or ChatGPT with the relevant connected apps. If a connection is unavailable, supply a calendar export, task list or sample data and keep it clearly labelled.
2
Second Brain
one folder + searchable answers
Find the document you need without remembering where you saved it.
This is a document library, not the business-memory brief in system #10 below. Start with a small folder of contracts, insurance, contacts, notes and PDFs. Make it searchable before you add the rest.
Help me build a searchable Second Brain from my documents.
My starting folder is [FOLDER OR UPLOADED FILES]. It may contain contacts, agreements, insurance, notes and PDFs. First tell me what you can actually read in this chat. If you cannot access the folder, ask me to upload a small test set. Do not claim to have read files you cannot open.
For the files you can read, make an index with: filename, document type, person or organisation, important dates, one-line summary and the original file location. Flag duplicates and missing information. Keep the originals where they are; do not rename, move or delete anything without asking me.
Then show me three questions I can ask this library, such as “When is my insurance due?” Answer each using only the documents, with the filename and relevant passage beside the answer. Say “I couldn't find it” when the source is missing.
Give me a simple repeatable way to add new files. If I use Obsidian, show me how this index could live in a vault; otherwise use a plain folder and index.
Test with five files first. Check every answer against the original document before trusting the library with the rest.
Works in:Claude or ChatGPT with uploaded files or supported file connections. Obsidian is optional; the prompt does not assume a direct integration.
3
Executive Assistant Agent
morning plan + evening check-in + weekly review
A useful check-in that remembers what actually happened.
Give this assistant a simple daily log. It can plan the day, ask what worked at night and spot patterns at the end of the week. The review should come from your answers, not guesses.
Help me build an Executive Assistant Agent for daily and weekly reviews.
Create one simple log I can keep using, with a date for each entry. Each morning, ask what is already on my calendar, what matters most today and what I am likely to avoid. Give me a short plan with no more than three priorities.
Each evening, ask me one question at a time: What did I finish? What slipped? What went well? What was frustrating? What needs to change tomorrow? Wait for my answer before asking the next question. Record my answers without polishing away the useful details.
At the end of the week, use only those dated entries to show: what moved forward, the repeated blockers, one thing to stop doing and three specific changes for next week. Quote the entries behind each pattern. If a day is missing, say so.
Start by building the log and running today's check-in with me. After I have used it manually, explain the reminder or scheduled-task options that are actually available in my account. Do not set them up until I approve the timing.
The nightly answers are the data. If you skip them, the weekly “analysis” is just a plausible story.
Works in:Claude or ChatGPT in one ongoing Project or chat. Reminders and scheduled runs depend on the features available in your account.
4
Content Marketing Agent
brand + customer + source material → usable assets
Make marketing drafts that start with your business, not generic AI ideas.
Feed it your actual offer, ideal customer and examples of your writing. Then make three different assets from one real source: a social post, an email and a simple visual brief.
Help me build a Content Marketing Agent for my business.
First, interview me for the minimum useful context: my offer, ideal customer, common objections, proof I can actually use, brand voice and examples of writing I have sent. Ask for real source material before drafting. Put the answers into a short brand-and-customer brief I can edit.
Then use [TRANSCRIPT, CUSTOMER QUESTION OR ORIGINAL IDEA] to produce three different assets:
1. A social post with one clear point and a strong first line.
2. A short email that teaches one practical thing and ends with one relevant action.
3. A visual brief with the headline, main image idea and three details the designer needs.
Do not repeat the same wording across all three. Do not invent testimonials, numbers, customer quotes or product features. Mark any claim that needs me to verify it. Show which source detail each asset uses.
Give me the reusable agent instructions and a quick review checklist. Save everything as drafts; nothing gets published or sent until I approve it.
The brand brief is the asset you keep. The three drafts are how you test whether the agent has actually learned it.
Works in:Claude Projects or ChatGPT Projects with your approved brand material. Publishing requires a separate review and connected workflow.
5
AI Workforce
chief of staff + specialist roles + clear hand-offs
Give each AI job an owner, then test one hand-off.
A “team of agents” sounds impressive until nobody knows who does what. Start with a chief of staff, three specialist jobs and a rule for when work comes back to you.
Help me design a small AI Workforce for [BUSINESS] around one real workflow: [WORKFLOW].
First, map the steps a person does today. Then propose a chief-of-staff role and up to three specialist roles. For each role, write: its job, the information it needs, what it produces, what it must never do, and when it hands work to the next role or back to me.
Give the chief of staff one intake checklist, a simple hand-off format and a final quality check. Keep private information within the tools I have approved. Nobody sends messages, publishes, spends money, changes customer records or makes a final customer promise without my review.
Now run one example with me using [REAL OR TEST TASK]. Show each role's output and the hand-off. If this app cannot run separate agents, use these as role briefs in one chat or Project and tell me plainly. Do not imply that an autonomous team is live.
After the example, tell me which steps could be automated with my available tools and which should stay human-reviewed.
One working hand-off is a better first build than five agent names and a diagram.
Works in:Claude or ChatGPT can draft and test the roles. Separate autonomous agents and external actions need supported tools, connections and explicit setup.
The 10 systems
Each one has the full prompt, a copy button, and where the wording matters most. Swap anything in [brackets] for your own details. You do not need to build all ten — see the next section.
1
Morning brief agent
calendar + inbox + news → one page
You open your laptop and the thinking is already done.
Before you've made coffee, Claude has read your calendar, your unread email and what's moved in your industry overnight — and turned it into one page. Not a summary of everything. A decision about what actually matters today.
You're my morning chief of staff. Give me ONE page — nothing longer.
Read my calendar for today and my unread email from the last 24 hours.
Then give me exactly this:
TODAY — every meeting, the time, who's in it, one line on what it's for
NEEDS ME — anything waiting on a decision or reply from me, most urgent first
CAN WAIT — things I should know about but don't need to touch today
ONE THING — if I only get one focused hour today, what should it go on?
No preamble. Start at TODAY. If something looks like it's about to become a
problem, say so plainly.
The magic isn't the summary — it's ONE THING. That line is what stops you opening your laptop and drifting into email for an hour.
Works in:Claude (Gmail + Calendar connectors, then save it as a 6am routine) · ChatGPT (connect the same apps, then set a scheduled Task) · Codex (best if your notes and to-dos live in files on your machine)
Watch Brooke build it — 3 Claude automations that run your morning for you
2
Prospect research pipeline
name → research → angle → ready-to-send
Stop sending the same message everyone else sends.
You give it a name and a company. It comes back with who they are, what they're actually dealing with this month, and three opening lines that reference something real — so you never have to write "I loved your recent post" again.
Research this person before I reach out: [NAME], [ROLE] at [COMPANY].
Search the web. I want:
WHO THEY ARE — role, how long, what they did before
WHAT'S HAPPENING — anything from the last 90 days: funding, launches, hires,
press, things they've written. Link every claim.
THE GAP — based on what they've said publicly, what problem are they probably
living with right now?
MY OPENER — three first lines. Each references something specific and real
from above. No compliments. Under 30 words each.
If you can't verify something, write "couldn't verify". Do not fill gaps with
plausible-sounding guesses.
That last line — "do not fill gaps with plausible-sounding guesses" — is the whole system. Without it you'll get a confident, invented paragraph about a funding round that never happened.
Works in:Claude (turn on web search) · ChatGPT (use deep research for the bigger accounts) · Works on the free plans too
No video for this one yet — it's on the list.
3
Content repurposing engine
one piece → many formats
One thing you already made becomes a week.
Most repurposing tools rewrite the same point five ways and it's obvious. This one pulls out five separate ideas first, then writes each format around a different one — so your week doesn't read like one post wearing five hats.
Here's one piece of content I've already made: [PASTE TRANSCRIPT OR LINK]
Turn it into a week — but pull a DIFFERENT idea out for each one. Do not
rewrite the same point five ways.
First, list the 5 separate ideas you found. If you can only find three, say so
— don't pad it.
Then write:
1. LinkedIn post — the argument, told straight
2. Instagram caption — the story or the moment, not the lesson
3. Three short-form hooks — the most surprising or contrarian bits
4. An email — one practical thing the reader can do today
5. A carousel outline — 7 slides, one idea per slide
Match my voice from the source. Don't add words I wouldn't say.
Making it list the five ideas before it writes anything is what stops the repetition. Don't skip that step.
Works in:Claude (save it as a Project so your voice and rules stick) · ChatGPT (build it as a custom GPT) · Codex (if your content lives in files, it can write the whole week to disk)
Watch Brooke build it — How I post 150 times a week as a solo founder
4
Meeting prep system
company + contacts + news → ready to call
Never open a call cold again.
Two minutes before you dial, you've got one page: who they are, what you promised them last time, what to ask, and what to say if it goes sideways. It reads your own email history, so it knows what you said you'd do — and whether you did it.
I've got a call with [NAME] from [COMPANY] at [TIME].
Pull together my prep:
THE COMPANY — what they do, size, anything that's changed recently
THE PERSON — role, what they care about, how they talk publicly
OUR HISTORY — search my email and calendar for every previous interaction.
What did we say we'd do? Did I actually do it?
3 TALKING POINTS — specific to them, not generic
2 QUESTIONS — the ones that actually move this forward
THE RISK — what could go sideways, and what to say if it does
One page. I'm reading this in the two minutes before I dial.
"Did I actually do it?" is the line that saves you. It catches the thing you promised six weeks ago and forgot.
Works in:Claude (Gmail + Calendar connectors — this is the one that makes it feel like magic) · ChatGPT (connect your email, or paste the thread in) · Set it as a routine that runs an hour before any call
Watch Brooke build it — Automate your client call prep with Claude
5
The decision council
different views → better decisions
Five advisors who argue. Not one that agrees.
The biggest problem with AI advice is that it tells you what you want to hear. This makes five advisors argue with each other about your decision — including one whose only job is to attack it — before giving you a verdict.
I need a real decision, not agreement. Here it is: [YOUR DECISION]
Convene a council. Five advisors, and they must genuinely disagree:
THE OPERATOR — only cares whether this can actually be executed
THE ACCOUNTANT — only cares about the money and the downside
THE CUSTOMER — only cares whether anyone actually wants this
THE SCEPTIC — argues against it. Hard. Find the real weakness.
THE STRATEGIST — cares where this leaves me in 12 months
Each one: 3 sentences max, in their own voice, and each must take a position.
Then THE VERDICT — what you'd actually do, the strongest argument against it,
and the one thing that would change your mind.
If the council mostly agrees, say so — and tell me this decision is probably
easier than I think.
This is the one to try first, because it needs zero setup. Paste it into whatever you already have open and give it a decision you're actually stuck on.
Works in:Claude · ChatGPT · Gemini — this one is pure prompt, so it works anywhere, on any plan, right now
Watch Brooke build it — Claude hallucinates up to 48% of the time — Claude Council fixes it
6
The money system
plan · track · save · grow
Where it actually went. Not where you think it went.
Export three months of transactions, hand them over, and get the honest version: every subscription you forgot, everything that quietly went up in price, and the three changes that free up the most money for the least pain.
Here are my last 3 months of transactions: [UPLOAD CSV OR PASTE]
Be blunt with me.
WHERE IT WENT — group everything into real categories, biggest first, with the
monthly average
THE LEAKS — every subscription or recurring charge. Flag anything I'm paying
for twice, anything that's crept up in price, and anything that looks
forgotten.
THE PATTERN — what does my spending say about how I actually run this
business, versus how I think I run it?
THE PLAN — the three changes that free up the most money for the least pain.
Give me the dollar figure for each.
Don't be nice about it. I want the number.
Ask for the dollar figure for each or you'll get advice instead of numbers. Brooke found $1,000 a year doing exactly this.
Works in:Claude (upload the CSV straight from your bank or accounting software) · ChatGPT (same) · Neither needs to connect to your bank — you're just handing over a file
Watch Brooke build it — I had Claude audit my subscriptions — here's the $1000/year it found
7
Your voice clone
your writing → trained → sounds like you
The difference between writing for you and writing like you.
Give it enough of your real writing and make it build a style guide — sentence shapes, the phrases you reach for, the things you'd never write. Then it stops producing that flat, tidy AI voice everyone can spot.
I'm going to teach you my writing voice. Here are 5-10 things I've actually
written: [PASTE THEM]
Don't summarise them. Analyse them and build me a style guide:
SENTENCE SHAPE — how long, how varied, where I break the rules
MY WORDS — the phrases I reach for again and again
WHAT I NEVER DO — constructions that appear nowhere in my writing
HOW I OPEN — the real pattern in my first lines
HOW I CLOSE — do I land it, ask something, trail off?
THE TELL — if someone read one paragraph, what would give it away as mine?
Then write one new paragraph on [TOPIC] using it. Then tell me honestly which
bits you got right and which still sound like AI.
Save this style guide to memory and use it every time I ask you to write.
WHAT I NEVER DO does more work than everything else combined. Telling it what to avoid beats telling it what to aim for.
Works in:Claude (save the style guide to memory or a Project so it sticks) · ChatGPT (paste the finished guide into custom instructions) · Codex (drop it in your project file and every session inherits it)
Watch Brooke build it — Ask Claude to humanise your text — it passed every AI detector
8
Weekly review loop
review + reflect + plan + repeat
Twenty minutes on Friday that changes the next week.
Not a journal prompt. An interview. It asks one question at a time, waits for your answer, and pushes back when you're vague — which is the entire point, because "I was busy" is not a reason.
It's my weekly review. Walk me through it — one question at a time, and wait
for my answer before moving on.
1. What actually moved this week? (make me be specific)
2. What stalled, and what was the real reason?
3. What did I spend time on that didn't matter?
4. What am I avoiding?
5. What are the three things that have to happen next week?
Push back on my answers. If I say I was "busy", ask with what. If my three
things for next week are vague, make me sharpen them.
At the end: the week in one honest sentence, and next week's three things
written so specifically I can't wriggle out of them.
One question at a time, and wait. Without that line it dumps all five at once and you write three lazy sentences instead of thinking.
Works in:Claude (set it as a routine for Friday afternoon so it comes to you) · ChatGPT (a scheduled Task does the same) · Keep it in one long-running chat so it remembers last week
Watch Brooke build it — Claude planned my whole week in 2 minutes (Calendar + Gmail)
9
Inbox triage
sort → draft → done
The boring replies are already written when you wake up.
Everything sorted into four piles, and every easy reply already sitting in your drafts folder in your voice. You're approving and sending, not writing from scratch. Nothing goes out without you.
Go through my unread email from the last 24 hours and sort it:
NEEDS ME — a real decision only I can make. Tell me what the decision is.
QUICK REPLY — I need to answer but it's not hard. Draft it for me.
FYI — I should know, no action needed
NOTHING — newsletters, receipts, noise
For everything in QUICK REPLY, write the actual draft. My voice: warm, direct,
short. No "I hope this email finds you well" — get to the point in the first
line. If I owe someone something late, acknowledge it without grovelling.
Put the drafts in my drafts folder. Do not send anything.
Keep "Do not send anything" in the prompt every single time. It's the line that lets you sleep.
Works in:Claude (Gmail connector, then a routine for 6am) · ChatGPT (connect Gmail and use a scheduled Task) · Both can write drafts without ever sending
Watch Brooke build it — I let Claude run my inbox for a week — here's the exact setup
10
Your second brain
all in one place
Stop explaining your business every single time.
One structured brief holding how you make money, who your clients are, your prices, and the decisions you've already made and don't want to argue about again. Once it's in memory, every answer you get is about your business.
I'm building you a memory of my business. Organise it — don't just store it.
Here's the raw material: [PASTE OR ATTACH — how you make money, who your
clients are, your prices, your tools, how you like to work, and decisions
you've already made and don't want to revisit]
Turn it into a structured brief with these sections:
THE BUSINESS · THE CLIENTS · THE OFFERS AND PRICES · THE TOOLS ·
HOW I WORK · SETTLED DECISIONS (things not to re-open)
Then tell me what's missing — what would you need to know to give me genuinely
useful advice that isn't in here yet?
Save this to memory and reference it every time. I shouldn't have to explain
my business again.
SETTLED DECISIONS is the underrated bit. It stops the AI reopening a pricing debate you finished eight months ago.
Works in:Claude (memory + a Project, and connect Drive so it can read your real documents) · ChatGPT (Projects and memory work the same way) · Codex (keep the brief as a file — every session picks it up automatically)
Watch Brooke build it — How to make Claude remember your whole business (3-layer memory setup)
Which one to build first
Nineteen prompts are a menu, not a weekend plan. Pick one of the weekend agents or projects above, or use this order for the quicker systems.
1
Right now, before you close this page — the decision council (#5)
Zero setup. Paste it into whatever you already have open and give it a decision you're genuinely stuck on. It takes two minutes and it's the one that makes you believe the rest.
2
Saturday morning — inbox triage (#9) or the morning brief (#1)
Pick whichever one describes your worst hour of the day. These two need your connectors switched on, so do them while you've got a bit of patience.
3
Sunday — your voice clone (#7)
Do this one before any of the writing systems. Every other system that writes anything gets better once it knows how you sound, so it quietly upgrades the rest.
Three is a good weekend
Build three, use them for a fortnight, then come back for the next three. A system you actually use beats nine you set up once and forgot.
Claude, ChatGPT or Codex?
All 19 prompts can start in Claude or ChatGPT. What changes is which files and connected tools your account can actually use, and how far you can automate them.
1
Claude — the one these were built for
Connectors read your Gmail, Calendar and Drive. Projects hold your instructions. Routines run a prompt on a schedule, so the morning brief arrives at 6am without you asking. This is the smoothest path.
2
ChatGPT — nearly everything, slightly different names
Connect the same apps, use Projects the same way, and use Tasks where the guide says routine. Custom instructions are where your voice guide lives.
3
Codex — only if your work lives in files
It runs on your own machine and reads your folders, which is brilliant for the repurposing engine and your second brain, and overkill for the rest. Your instructions live in a project file every session picks up automatically.
Make them run without you
A prompt you have to remember to paste is a chore. A prompt on a schedule is a system — and that's the whole difference.
Once a prompt is giving you a good answer consistently, save it as a routine in Claude (or a Task in ChatGPT) and give it a time. Five of these are worth putting on a timer:
Morning brief — every weekday at 6am
Inbox triage — 6am, and again after lunch if your inbox is busy
Meeting prep — one hour before anything in your calendar
Weekly review — Friday afternoon, so it catches you before you switch off
Competitor watcher — Sunday afternoon, so your content ideas are ready that night
Give it a fortnight before you automate
Run a prompt by hand for two weeks first and tweak the wording as you go. Automating a prompt that isn't quite right just means being wrong on a schedule.
Want to build these with me?
Inside the Wright Mode membership I build these live, share the ones that actually stuck, and troubleshoot yours when they don't behave. Come and have a look →