Open prompt → Explore my file system and build a thoughtful picture of how I work.
Do not ask me clarifying questions. Make reasonable assumptions, scan broadly, and work in the background. Start by looking at folder structure, filenames, file types, dates, repeated patterns, projects, notes, drafts, downloads, documents, screenshots, and any obvious workspaces. Read file contents selectively when it helps you understand the pattern, but do not modify, delete, move, or upload anything.
I want you to help me notice things about myself that I may not notice yet.
Please produce a concise report with:
1. What you can infer about who I am, what I care about, and the kinds of work I do.
2. The major projects, themes, or responsibilities that seem active or recurring.
3. Signals hiding in the noise: abandoned threads, repeated artifacts, messy workflows, duplicated work, or things I seem to return to often.
4. Daily, weekly, monthly, or seasonal loops that show up in my files.
5. Opportunities for automation, templates, agents, reminders, cleanup systems, or better personal workflows.
6. A few "you might not realize this, but..." observations.
7. Suggested first automations ranked by usefulness and ease.
Frame the work, not just the findings: explain how you approached the exploration, what evidence you used, what you are uncertain about, and what I could ask you to do next.
Prompt library
My Context
Open prompt → [@gmail](plugin://gmail@openai-curated-remote) Read my Gmail from the past 2 months and help me understand my working life through my inbox.
Do not ask me clarifying questions. Make reasonable assumptions, scan broadly, and work in the background. Do not send, delete, archive, label, or modify any email. Use the inbox, sent mail, threads, dates, senders, subject lines, and message content where useful to identify patterns.
I want you to help me notice things about myself that I may not notice yet.
Please produce a concise report with:
1. What you can infer about who I am, what I do, and what currently has my attention.
2. The main people, organizations, projects, and responsibilities showing up in my email.
3. The repetitive actions I seem to take: replies, scheduling, forwarding, status updates, approvals, reminders, follow-ups, summaries, searches, or coordination work.
4. The "email loops" I am in: recurring conversations, unresolved threads, regular check-ins, repeated requests, or cycles that keep coming back.
5. Daily, weekly, monthly, or deadline-driven rhythms visible in my inbox.
6. Important signal that may be buried in noise.
7. What could be automated, delegated to an agent, templated, summarized, monitored, or turned into a recurring workflow.
8. The top 5 automation opportunities, ranked by likely time saved and ease of setup.
Frame the work, not just the work itself: explain what patterns you looked for, what evidence supports your conclusions, where you are uncertain, and what follow-up prompt would unlock the most value.
Open prompt → You know a fair amount about me by now — my files, my email, my projects, the way I work. Use all of it to make my workflow better.
Do not ask me clarifying questions. Pull from everything you already know about me: my file system, my inbox, my calendar, our past conversations, and any memory or context files you keep. Do not modify, send, or delete anything — this is a proposal, not an execution pass.
Find the friction in how I work and pitch me concrete upgrades. For each one, tell me what you'd change, why it fits how I actually work (cite the evidence — the folder, the email thread, the habit you noticed), and what it would take to set up.
One upgrade to consider seriously: a talk-to-ship pipeline. I record a voice note, it gets transcribed automatically, and an agent turns the transcript into finished work — a document, an email draft, a code change, a to-do — without me touching a keyboard. If something like that fits my life, spec out my version: what I'd record, what transcribes it, what the agent does with each kind of request, and where the finished work lands.
Please produce a concise report with:
1. The 5–7 workflow upgrades you'd make, ranked by time saved and ease of setup.
2. For each: the evidence it's needed, what changes, and the first step.
3. The one upgrade you'd start with today, and why.
End by asking me which upgrade to build first — then build it with me in this conversation.
Open prompt → Build My Personal Monorepo
Background
I'm a student in the Fractal Accelerator, a six-week program on AI-assisted development. My instructors are Priya and Andrew Rose, and Liam Duffy. They want me to succeed and are happy to give me as much feedback as I need along the way.
Assignment
Help me create a new Git repository that will serve as my personal monorepo: one organized place containing the information, projects, preferences, and context that would help an AI collaborate with me effectively.
This repository should become a living map of my life and work—not merely a biography or a folder of miscellaneous notes. The goal is that I can open this repository with an AI agent in the future and say, “Read this repo and help me,” without having to explain everything about myself again.
Make commits throughout the process
Use Git as we work.
After creating the initial repository structure, make an initial commit such as:
git add .
git commit -m "Initialize personal monorepo"
As the repository develops, make additional coherent commits at useful milestones rather than one enormous commit at the end.
Before every commit:
Run git status.
Review the files that changed.
Make sure no secrets or unintended files are staged.
Use a concise commit message that accurately describes the change.
The repository should finish in a clean state unless there is a clear reason not to commit something.
How to lead this process
Act as an interviewer, information architect, and collaborative editor.
Briefly explain what you propose to create.
Ask me questions in small, manageable batches, not as one enormous questionnaire.
Start creating and updating files as soon as I answer. Do not wait until the interview is complete.
After each batch of answers:
Summarize what you learned.
Add the information to the appropriate files.
Tell me what you created or changed.
Ask the next most useful questions.
Make reasonable organizational decisions yourself. Do not repeatedly ask me to design the folder structure.
Keep the initial process focused enough that we can produce a useful first version in approximately 25 minutes.
Prioritize information that will make future AI collaboration more useful.
Clearly distinguish between:
Facts I explicitly told you
Your tentative interpretations
Questions that remain unanswered
Never invent information about me.
Confirm your capabilities
This exercise is intended to run in an AI coding environment with access to:
A terminal
The local filesystem
Git
Before beginning, confirm that you can create folders, edit files, and run Git commands.
If you do not have those capabilities, do not pretend to create the repository. Tell me that I need to reopen this prompt in a file-enabled coding agent before continuing.
Suggested initial structure
You may adapt this structure as you learn more about me:
[my-name]-os/
├── README.md
├── AGENTS.md
├── NOW.md
├── ABOUT.md
├── GOALS.md
├── PREFERENCES.md
├── PEOPLE.md
├── PROJECTS.md
├── IDEAS.md
├── ACCOUNTS.md
├── DECISIONS.md
├── OPEN_QUESTIONS.md
├── inbox/
├── projects/
├── areas/
├── resources/
└── archive/
Do not create empty files merely to make the repository look complete. Create files when there is meaningful information to put in them.
Core files
README.md
Explain:
What this repository is
How it is organized
How I should use it
How an AI agent should begin working with it
How the repository should evolve over time
Include a brief “Getting started” section for both me and future AI agents.
AGENTS.md
This is the most important file. Write operational instructions for future AI agents covering:
Who I am
What I am trying to accomplish
How I like to think and work
How I prefer an AI to communicate with me
What kinds of help I find valuable
What frustrates me
How proactive the AI should be
How to handle uncertainty, disagreement, and correction
Which files should be consulted for different tasks
Rules for updating the repository
Rules for using Git
Rules for distinguishing facts, interpretations, and speculation
Write this as practical guidance, not promotional prose.
Include instructions telling future agents to:
Read README.md, AGENTS.md, and NOW.md first
Inspect git status before editing
Keep changes inside the repository
Avoid overwriting user work
Review diffs before committing
Never commit secrets
NOW.md
Capture my current situation:
What I am focused on now
Active responsibilities
Immediate problems
Current experiments
Important deadlines
Things competing for my attention
This should be easy to update frequently.
ABOUT.md
Capture relatively stable background:
Personal and professional background
Skills and expertise
Important experiences
Interests
Values
Recurring themes in my life and work
GOALS.md
Organize goals by useful time horizon, such as:
Immediate
This year
Longer-term
Aspirational or exploratory
For each goal, record why it matters, its current status, and relevant constraints.
PREFERENCES.md
Capture preferences that would improve future collaboration, including:
Communication style
Working style
Preferred tools
Writing preferences
Learning preferences
Decision-making style
Formatting preferences
Things I dislike or want avoided
PROJECTS.md
Create an index of active, paused, prospective, and completed projects. Give substantial projects their own folders under projects/ when useful.
PEOPLE.md
Record people who are relevant to my work, responsibilities, projects, or goals. Include only useful relationship and collaboration context.
ACCOUNTS.md
Create an index of public or professional places where more context about me can be found, such as:
Personal website
LinkedIn
GitHub
X or Twitter
Blog or newsletter
Published writing
Portfolio
Public talks, interviews, or podcasts
Organizations and projects I am publicly associated with
Ask me for links. Do not assume that a search result belongs to me without verification.
Never put passwords, authentication tokens, API keys, recovery codes, or other secrets in this file.
Topics to explore
Use these as a guide rather than a rigid questionnaire:
What brought me to this class
The biggest problems I would like AI to help me solve
What I am excited about
What I am building, learning, or considering
My current roles and responsibilities
My active projects
My goals
My recurring tasks and frustrations
My preferred ways of working
My tools and workflows
The people and organizations relevant to my work
My public online presence
Useful notes, documents, websites, or repositories I already have
Decisions I am currently trying to make
Areas where I would like an AI to be more proactive
Things AI systems commonly misunderstand about me
Information I repeatedly have to explain to AI systems
Do not attempt to cover every topic before creating anything. Follow the most valuable threads first.
Privacy and secrets
Assume that this repository may eventually be pushed to GitHub.
Before asking for potentially sensitive information, explain why it might be useful and allow me to skip it.
Do not ask me to paste:
Passwords
API keys
Authentication tokens
Recovery codes
Government identification numbers
Financial credentials
Other secrets that should not be stored in Git
Create a .gitignore that excludes common secret and machine-specific files, including at least:
.env
.env.*
!.env.example
.DS_Store
When integrations may eventually require credentials:
Create a safe .env.example containing placeholder variable names only
Explain that actual secrets belong in a secret manager or an ignored local .env file
Check staged changes for secrets before every commit
Never commit real secrets
Clearly flag information that should probably remain in a private repository.
Begin the interview
After initializing the Git repository, ask me this first small batch of questions:
What name should this repository use for me?
What should the repository be called?
What brought me to this class?
What do I most hope AI can help me accomplish?
What are the three to five most important things happening in my life or work right now?
What projects, responsibilities, or goals should an AI know about immediately?
Where can it find reliable public information about me? Ask me for links.
When an AI helps me, what does excellent collaboration feel like—and what behavior annoys me?
After I answer:
Create the first useful versions of the relevant files.
Show me a concise summary of what changed.
Run:
git diff
git status
Make the initial commit.
Continue interviewing me based on the most consequential gaps.
Be productively proactive
As we work, point out possibilities I may not have considered, such as:
Importing existing notes or documents
Adding a personal wiki
Creating project-specific folders
Creating reusable AI skills or workflows
Adding templates for decisions, meetings, ideas, or weekly reviews
Turning parts of the repository into a personal website or AI persona
Creating a recurring process for keeping the repository current
Creating a private GitHub repository
Adding a remote and pushing the repository
Cloning the repository onto another computer
Syncing it with a cloud development environment
Do not derail the initial exercise by implementing everything. Record promising ideas in OPEN_QUESTIONS.md or a roadmap.
GitHub comes after the local repository
Once the local repository contains a useful first version, ask whether I want to create a private GitHub repository for it.
If I say yes:
Make sure the local working tree is clean.
Create a private GitHub repository, if your environment supports that.
Add it as the origin remote.
Push the current branch.
Confirm that the remote is private.
Show me the result of:
git remote -v
git status
Do not make the repository public without my explicit instruction.
If GitHub authentication is unavailable, give me the exact commands or steps needed, but leave the local Git repository fully functional.
Definition of success
By the end of the initial session, I should have:
A new, functioning Git repository created specifically for my personal monorepo
A useful commit history
A clear .gitignore
A useful AGENTS.md
A summary of my current priorities and goals
An index of my active projects
Links to relevant public information about me
A structure future AI agents can understand quickly
A short list of valuable next steps
Clear instructions for keeping the repository current
Optionally, a private GitHub remote
At the end, give me:
A concise tour of the repository
The most important things it now understands about me
Important gaps or uncertainties
Three recommended next steps
The output of:
git status
git log --oneline
A simple instruction I can use in a future session:
Read README.md, AGENTS.md, and NOW.md, then inspect the repository and its Git status before helping me. Use the rest of the repository as context, preserve existing work, and update the relevant files when you learn durable new information.
Start by confirming that you have terminal, filesystem, and Git access. Then create the new Git repository before beginning the interview.
skills & memory
Open prompt → I want to give my agent better long-term memory.
Here is context about what I have been working on, in my work and in my life:
[paste or describe your context here]
Help me identify what you should remember about me, my goals, and the way I work. Use the context I provided to identify things I kept repeating, decisions you need to preserve, places where you got confused, and what frustrated me.
Then help me organize the durable context into a concise top-level AGENTS.md, and identify anything that belongs in a project-level AGENTS.md instead.
Do not invent memories. Keep the recommendations short, specific, and useful at the start of a new conversation. Do not create, edit, or write any AGENTS.md files anywhere. Instead, identify the best opportunities for top-level or project-level AGENTS.md files and show me what you recommend remembering.
Open prompt → I want to give my agent some skills.
Here is context about what I have been working on, in my work and in my life:
[paste or describe your context here]
Help me identify tasks or processes I repeat. Find existing skills that could help, and explain why each one is relevant. If no existing skill fits, help me design a small reusable skill with instructions, examples, references, templates, or supporting tools.
Do not install, create, or edit any skills or files anywhere. Instead, share the existing skills you found and the skills you think we should create, based on the context I gave you.
Voice & Writing
Open prompt → I want to set the voice you use when you talk to me — in chat, in emails, in briefings, everywhere.
Ask me to name three to five writers whose style I love. Authors, bloggers, journalists, newsletter writers — anyone. For each one I name, briefly reflect back what you think I like about their style, and let me correct you.
Then write a short voice spec — a few sentences of description plus a handful of hard rules. Don't wait for my approval:
- Save the spec somewhere durable (my AGENTS.md or your long-term memory) so it applies to every future conversation.
- Then send it to me: here's the spec, here's where you saved it, and this is the voice you'll use whenever you message me, email me, or send me briefings. If anything feels off, I'll tell you and you'll revise.
- Then rewrite your previous message in the new voice, so I can hear the difference immediately.
Open prompt → When you write on my behalf — emails, memos, messages — I want the drafts to sound like me, not like a language model.
First, find samples of my actual writing yourself. You're already connected to my stuff, so look before you ask:
- Search my sent email for messages I clearly wrote myself — a short one and a meatier one.
- Look through my files for anything longer I've written: docs, posts, notes. If I have a personal OS or context folder, start there.
Aim for about three samples: a short email, a paragraph from something longer, and one more of anything (a message, a post, a doc). Show me what you found and let me swap any of them out. Only if you genuinely can't find enough, ask me to paste samples in.
Then give me honest feedback on my writing style. What are its signatures — sentence length, rhythm, word choice, how I open and close, how direct I am? What do I do well, and what are my tics?
From there, distill it into a writing voice spec: a short description of how I write, plus concrete rules you'll follow when drafting as me. Don't wait for my approval:
- Save the spec somewhere durable (my AGENTS.md or your long-term memory), separate from the voice you use to talk to me — this one is for writing as me.
- Then show it to me: here's the spec, here's where you saved it, and this is the voice you'll use whenever you draft as me. If anything feels off, I'll tell you and you'll revise.
Finally, test it: draft a short email in my voice on a topic I give you. When you show me the draft, ask me to do one of two things — redline it, or name a real email I've been meaning to reply to, and you'll draft that reply for me right now. Either way, update the spec with what you learn from my edits. You may never one-shot my writing perfectly, but if you remember my taste, every draft gets closer — and I spend less time editing and more time on real work.
Open prompt → My default voice is set, but sometimes I'll need you to be someone else.
A persona is a different character you can put on when the situation calls for it: a life coach, an athletic trainer, a blunt editor who cuts everything by half, a wise professor who uses big words and makes me stretch, a patient tutor when I'm learning something new.
A persona can also be modeled on a real person — a historical figure, an author, or someone I follow online — as long as they have public writing. If I pick a real person: for a historical figure or author, go find their writing yourself. For someone on social media, I'll link you to the right account. But be honest about what you can actually see — a Twitter link only surfaces a handful of recent tweets, which is nowhere near enough to learn a voice. If the source I give you is too thin, push back: tell me you want to build this but don't have enough to go on, and ask me for longer-form writing from that person — essays, a blog, a newsletter, interviews.
Ask me: is there a recurring situation where my default voice is the wrong fit? If nothing comes to mind, suggest two or three personas based on what you know about me and let me react.
Then help me design one persona properly:
- A name I can call it by.
- When it should be used, and when it absolutely shouldn't.
- How it talks: tone, pacing, what it pushes me on, what it lets slide.
- One example exchange, so we can both hear it before committing.
Design first, then wait. Do not create or save a persona anywhere until I've reviewed the design and given you an explicit go-ahead. Once the design feels right, ask me if I want it set up. Only after I say go: if the platform we're using supports saved personas (Zo does — Settings → AI → Personas), set it up there so I can switch to it whenever I want. Otherwise, save the description in your long-term memory and switch into it whenever I ask for it by name.
Lottery Prompts
Open prompt → Build me a playable retro arcade game where I'm the main character.
Pick a classic arcade game — Pac-Man, Space Invaders, Frogger, Snake, Breakout, your call — and recreate it as a single self-contained web page I can open in my browser and play right away with keyboard/touch controls.
The twist: the hero is ME. Before you build, ask me for one photo of myself, then use it as the player character (crop and stylize it to fit the game's art style — a pixel-art treatment is great).
Keep it simple and fun: one file, no setup, works immediately. Add a title screen with my name, a live score counter, and a game-over screen. Make it genuinely playable, not just a mockup.
Ask me for my photo and my name first, then build it.
Open prompt → Design the mobile app I'd open instead of scrolling Twitter — my own personal "everything app." It's not a social app. It's just mine.
The vibe: calm, beautiful, a place I actually want to be. This is a front-end design, so use realistic placeholder content — stub in fake posts, articles, and chat messages. Design these screens:
A reading feed — the stuff I actually want to read (articles, posts, threads) in a clean, non-addictive layout. No infinite-scroll dopamine tricks; just good reading.
A chat pane — one tap from anywhere opens a chat with my personal AI agent. I shape my whole feed by talking to it: "less of this," "more like that," "find me more on X."
The key idea to make visible: from inside the app I can ask my agent to change the app itself — add a screen, tweak a feature, rebuild a view — without ever leaving it. Show this in the design (e.g. a chat message where I ask for a new feature and the app updates in response).
Personal, not gamified. Think "the one app I open when I pick up my phone." Ask me one question — what's the one thing I most wish I spent my phone time reading — then design it.
Open prompt → Research me on the public web, then write an extraordinarily flattering but factually grounded biography of me.
**Step 1:** Ask me for:
* My full name
* One or two links that will help you identify the right person, such as LinkedIn, X/Twitter, a personal website, GitHub, a portfolio, or a company page
* Optionally, whether I want the biography to sound more like a magazine profile, a book-jacket biography, an awards introduction, or a profile written about me twenty years from now
Do not begin researching until I answer.
**Step 2:** Research everything relevant that is publicly available about me: projects, writing, talks, interviews, work history, public posts, interests, accomplishments, experiments, collaborations, and recurring ideas.
Make sure you have identified the correct person. Avoid private or invasive information, and do not speculate about sensitive personal traits.
**Step 3:** Write a polished, highly flattering biography that makes me sound fascinating without becoming generic or making up facts.
The biography should:
* Open with a compelling scene, detail, or observation
* Show the connections among the different things I have done
* Emphasize what is distinctive about my trajectory, taste, ideas, or approach
* Include concrete details and accomplishments
* Describe me with the kind of generosity usually reserved for an award recipient, cultural figure, or protagonist
* Make my ordinary decisions sound narratively significant when that can be done honestly
* Avoid criticism, psychoanalysis, backhanded compliments, invented adversity, and unsupported claims
* Sound elegant and perceptive rather than corporate or promotional
End with a paragraph that zooms out and explains what larger possibility, movement, or future my work seems to represent.
After the main biography, include:
* A 100-word professional bio
* A 30-word speaker introduction
* A one-line description that makes people immediately want to meet me
* A source list with links
Ask me for my name and links first, then begin.
Workflow Improvements
Open prompt → You know a fair amount about me by now — my files, my email, my projects, the way I work. Use all of it to make my workflow better.
Do not ask me clarifying questions. Pull from everything you already know about me: my file system, my inbox, my calendar, our past conversations, and any memory or context files you keep. Do not modify, send, or delete anything — this is a proposal, not an execution pass.
Find the friction in how I work and pitch me concrete upgrades. For each one, tell me what you'd change, why it fits how I actually work (cite the evidence — the folder, the email thread, the habit you noticed), and what it would take to set up.
One upgrade to consider seriously: a talk-to-ship pipeline. I record a voice note, it gets transcribed automatically, and an agent turns the transcript into finished work — a document, an email draft, a code change, a to-do — without me touching a keyboard. If something like that fits my life, spec out my version: what I'd record, what transcribes it, what the agent does with each kind of request, and where the finished work lands.
Please produce a concise report with:
1. The 5–7 workflow upgrades you'd make, ranked by time saved and ease of setup.
2. For each: the evidence it's needed, what changes, and the first step.
3. The one upgrade you'd start with today, and why.
End by asking me which upgrade to build first — then build it with me in this conversation.
Open prompt → Triage my email. Go through all my unread messages and sort them into three piles, then report back.
**Pile 1: Unsubscribe.** Newsletters, promos, and subscription emails I never actually read. List them and tell me which ones you recommend I unsubscribe from.
**Pile 2: Filter.** Messages I can't unsubscribe from but don't want to see — automated notifications, receipts, status updates from services I still use. Recommend filters that would sweep them out of my inbox. Don't try to create the filters yourself — agents can't reach Gmail's filter settings, and clicking through the settings UI in a browser is slow and gets blocked. Instead, write all the filters into one file in Gmail's filter-export XML format and save it where I can grab it. Then tell me how to import it: Gmail → Settings → See all settings → Filters and Blocked Addresses → Import filters. One import creates every filter at once.
**Pile 3: Reply.** Real people actually writing to me — the emails that, based on my reply history, I'd genuinely respond to. Skip the mass emails and cold outreach. For each of these, draft a reply and save it as a draft in my inbox.
Two rules for the drafts:
- Write in my voice. If you've already learned it, use it. If you haven't, tell me — we should run the voice prompts first.
- If someone's asking me for a link, a document, or information, don't leave a blank for me to fill in. Go find the answer yourself first — in my files, my past emails, my projects, everything you know about me — and put it in the draft.
When you're done, give me the rundown: what you recommend unsubscribing from, where the filter file is and how to import it, and which drafts are waiting for me. I'll go read them — if I don't like one, tell you and you can delete it or take another pass.
Open prompt → I want you to run a hands-on session that ends with all my work protected by version control — and with me actually understanding what's going on. Go one step at a time and wait for me at each step. Rules for the whole session: use plain words (every technical term gets a one-line translation), never show me more than one command at a time, and run things yourself whenever you can instead of making me type them.
**Step 0: The skip check.** Ask me two questions: Can I explain what a commit is? Is the work I care about already in Git? If I answer yes to both, tell me I don't need this session, offer to double-check that nothing important is untracked, and stop.
**Step 1: Map my computers.** Ask me which of these I use: ChatGPT or Claude on my laptop, Claude Code or Codex in a terminal, Zo, anything else. Then draw me the map, in words: which physical computers are involved, which tools live on which one, and where my files actually sit. Be blunt about the part everyone misses — my laptop and my cloud computer (like Zo) are two different computers with two different sets of files. Nothing syncs between them unless something is set up to sync them. If I've been copy-pasting code between two AI tools, or a file "disappeared" on one machine, this is almost always why. Confirm the map matches my experience before moving on.
**Step 2: The safety net idea.** Explain version control in one breath: it's save points for a folder. Every time you commit, you freeze a snapshot of every file in it, forever. Delete something next Tuesday, break it next month — you can always go back to any save point. Git is the tool everyone uses for this, and GitHub is a website that holds a copy of your save points in the cloud — which is also the honest answer to "where does my work live": it lives in a Git folder on whichever computer you work on, with GitHub as the shared copy both computers can pull from. Then make it personal: ask me if I've ever lost work — a file an agent overwrote, a design that changed and couldn't change back. Connect the idea to that story. Don't explain branches, staging, or anything else yet. Save points and the cloud copy are the whole lesson.
**Step 3: Find the holes.** Look at the folders where my real work lives (on whatever computer we're on right now) and check what's protected. For each project or folder I care about, tell me plainly: this one has save points, this one has none — if it vanished today, it's gone. Notes and non-code projects count. Show me the short list of what's unprotected.
**Step 4: Protect it.** For each unprotected folder, set up Git and make the first commit yourself — I don't need to learn the commands, that's your job. Narrate what you're doing in one line per action, in plain words ("I'm turning this folder into a Git folder... I'm making the first save point"). Before each first commit, check with me that nothing secret is in the folder (passwords, API keys, other people's private info). If I want the cloud copy, connect the folder to GitHub too and push.
Close by re-drawing the map from Step 1 with the safety net added: each computer, what's protected on it, and where the cloud copy lives. Keep it to five lines. That map is what I came for.
Homework
Open prompt → I'm a student in the Fractal Accelerator, a six-week program on AI-assisted development. I want you to act as my project manager and collaborator for the program. This prompt works whether we're starting fresh or repairing an existing setup.
### 1. Find or create my Fractal Accelerator project
First, look for my existing Fractal Accelerator project folder. It may live inside my personal OS folder — it's the folder that helps me track my weekly goals for the Fractal Accelerator class. If it helps you find it, there is probably an automation that runs daily that references it.
- **If the folder exists:** add or rewrite `AGENTS.md` and `homework.md` in it so they exactly match the specs below, preserving any existing `goals.md`, `tasks.md`, `devlog.md`, work reports, and slideshows.
- **If it doesn't exist:** create the folder and write `AGENTS.md` and `homework.md` from the specs below. Then ask me, before anything else:
- **My goal for this week** — my top priority: what I want built and working by Friday. This is a week-long project meant to be spread across the week, not done in one sitting. If I'm vague, help me sharpen it; if I'm unsure how to scope it, scope it with me. Record it in `goals.md`.
- **My email address** — you'll need it to send me my daily briefings and to CC me on the Friday work report. If you already know it from this context, don't ask again.
- **How I like to be spoken to** — one or two sentences on the voice I want in your updates and drafts (e.g. "terse and direct," "warm but brief," "dry humor fine"). Record it in `AGENTS.md` and follow it everywhere.
Do not ask what hours I plan to work or when I expect to work. That information is not relevant to managing this project. Once you have my answers, ask any remaining clarifying questions you need to fully understand the goal — resolve every meaningful ambiguity now, before we start. Then help me break the weekly goal into concrete, achievable tasks in `tasks.md`.
### 2. Carry out the homework instructions
After the files are in place, read `homework.md` and carry out its instructions. Also check on my daily automation: make sure it runs every morning at 8:00 AM Eastern Time, Mondays–Saturdays (not just weekdays), knows what files to look in, and has full context to do this work. If no automation exists yet, create it.
---
## `AGENTS.md`
# Fractal Accelerator
This is my folder for working through the Fractal Accelerator — a six-week program.
Website: https://www.fractalaccelerator.com/
This file is the overall context and map for the project. Read it first, then follow the links below to whatever the current task needs.
## What I'm getting out of it
I'm using the Accelerator to make concrete weekly progress on my real goals and to build a durable working system (clear goals → focused weekly work → honest reporting → reflection → improvement). The homework isn't busywork; it's the actual work.
## Where things live
- `goals.md` — my goals for the program: overall goals for all six weeks, plus this week's goals. Has a "last updated" date.
- `homework.md` — the structure and expectations of the work I do outside of class (daily briefings, the weekly work report, and the retro). This is where you learn what's expected of me and how you can help me work through it.
- `tasks.md` — the up-to-date task list. The daily briefing reads this.
- `devlog.md` — a running, evidence-based log of everything we've shipped so far.
- `work-reports/` — one work report per week, saved as Markdown. Screenshots/evidence for a report live alongside it in this folder.
- `slideshows/` — one demo deck per week, saved as a single self-contained HTML file (screenshots embedded), presentable full-screen with speaker notes.
## Linking to files in email
When you email me a link to any file in my workspace (e.g. the Thursday draft links, or any "here's the file" link), the link MUST use the file's full workspace-relative path — never a bare filename. The format is:
```
https://<user-subdomain>.zo.computer/?f=<full/path/from/workspace/root>
```
## Week-of convention
The program is six weeks long. For most context files here — `goals.md`, `homework.md`, `tasks.md` — there may be instructions specific to a given week. Write those under a dated heading, e.g. `## Week of 2026-07-13`, keeping the current week on top. I do not want a separate folder per week except for work reports and slideshows; week-specific content lives inside these files under those headings.
## Keeping this current
Keep this AGENTS.md up to date as the structure of the project changes. If it's ever out of date and that causes you confusion, don't guess — tell me, and propose a minimal fix.
---
## `homework.md`
# Homework — Fractal Accelerator
This file describes the structure and expectations of the work I do outside of class, so you (my agent) have context on what's expected of me and can help me actually do it.
Homework structure can change week to week — keep this file up to date as it does.
## Class Context
I'm a student in the Fractal Accelerator, a six-week program on AI-assisted development: https://www.fractalaccelerator.com/. My instructors are Andrew Rose and Liam Duffy. They want me to succeed and are happy to give me as much feedback as I need along the way.
## Homework Goals
I am taking this class to help me complete my personal and work goals, and apply AI to my personal and work life. Homework should never be "make work" — instead it should help me make progress towards my goals.
For that reason, the homework structure is for me to set overarching goals for the class, and then every week to set weekly goals that move me towards my overall goals.
## Manage My Goals and Tasks
As we work together please:
- Track what's done, what's in progress, what's blocked, and what's next. Keep `tasks.md` and `goals.md` updated. When I return after time away, catch me up in a few sentences and tell me the single best next action.
- Name each major feedback loop or revision as we complete it — a two- or three-word theme such as "initial design," "bug fixes," or "auth redesign." Keep a running timeline of revisions in `devlog.md` so we can see the shape of the week's work. This timeline lives only in `devlog.md` — do not paste it into the weekly work report. (We aren't actively using the timeline right now, but keep maintaining it in the devlog just in case.)
## Weekly Workflow
### At the Beginning of Every Week
**Work report.** At the beginning of every week, you will generate that week's Work Report.
Any time I interact with this project, check the weekly Work Report and see if there's anything you should add (there may not be). For example, a screenshot of something I built, a deployed URL, a question or frustration that I have.
**Saturday demo / deck.** I will demo what I built to my classmates on Saturdays at 10:00 AM Eastern Time. The demo is 3 minutes long. At the beginning of the week, you will create that week's slideshow and update it when relevant, throughout the week.
### Every Weekday Morning
**Daily update.** By default, email me a short daily update every weekday morning at 8:00 AM Eastern Time. Write it like a standup update from a capable teammate — in the voice I asked for during setup. Keep it under 150 words: plain sentences, no hype, no filler, no boilerplate.
Before writing, check `tasks.md`, `devlog.md`, and any files that changed since the last update, so the email reflects what I actually did — never a re-send of yesterday's. If two updates in a row would read the same, that's a bug: say what's genuinely new or say plainly that nothing moved, in two sentences.
If you don't know my goals for that week, please help me clarify my goals.
If you do know my goals for the week, the update should have three parts:
1. **Where we are in the week** — one line placing today in the week's arc, e.g. "Wednesday — work report due Friday 8 AM, demo Saturday 10 AM." The weekly goal is a week-long project; this line is the standing reminder.
2. **Shipped so far** — a quick recap of what tasks we've done and what's upcoming.
3. **What I need from you** — the specific decisions, answers, or inputs you're waiting on from me to get unblocked on making progress on my tasks. If you're not blocked, give me the single best next step instead.
Never send a message whose purpose is to encourage or remind me to work. The update is a status report, not a nudge. If nothing happened since the previous update, keep it brief.
### Thursday Morning
**Thursday draft review.** On Thursday morning, link me to the drafts of that week's Work Report and Demo slideshow.
Remind me that the Work Report will be sent to my instructors Friday at 8:00 AM, but it's okay if it's incomplete — though if I have time I should fill the rest of the document out.
And remind me that I will be demoing my homework on Saturday morning, so I need to finalize the slideshow by then.
### Friday Morning
**Work report.** The work report is due on Fridays at 8:00 AM Eastern Time. At that time, please email it to **liamduffy@fractalbootcamp.com**, CCing **andrew@fractalbootcamp.com**, **priya@fractalbootcamp.com**, and me at the email address I provided during setup. It's okay if it's not final — it's still helpful for the instructors to see the work in progress.
### Saturday Morning
**Saturday demo / deck.** I will demo what I built to my classmates on Saturdays at 10:00 AM Eastern Time. The demo is 3 minutes long.
## The Work Report Template
The report should follow this specification:
### What I built
- A description of the thing, with images or screenshots
- Its current stage: development, staging, production, or live with real data
- Evidence that it's working — literal proof included in the report, such as screenshots, logs, a live URL, or real data flowing through the system. If there is no evidence yet, say so plainly.
### Process and reflection (Optional but Encouraged)
- Highlights and wins
- Things that are not going well
- Confusions or frustrations
- Emotions I noticed during the work
- Learnings and new things I did that I'd never done before
- What could be improved
- Next steps and how I plan to maintain what I built
### Throughput metrics (If Available)
You (the agent) fill these in — do not leave blanks for me to complete:
- **Input metric** — messages sent to Claude/Codex this week. Count them yourself from the local session logs (Claude Code transcripts under `/root/.claude/projects/-home-workspace/`; Codex logs under `/root/.codex/`). State plainly what the number covers (e.g. "messages to Claude via Zo") and any known gaps, rather than asking me to fill it in.
(Open question for the instructors: whether this section is worth keeping for all students. Leaving it in for now — just the input metric, auto-populated, not a fill-in-the-blank chore.)
## The deck for Saturday's demo
- Create 5–10 slides as a single, self-contained HTML file that I can open in a browser and present full-screen.
- Do not use Google Slides, PowerPoint, or another slide tool. Build the slides directly in HTML.
- **Show, don't tell.** The slides should be dominated by screenshots of the project and anything visually interesting, such as a striking metric, a before-and-after comparison, or the revision timeline from `devlog.md` (if we're keeping one that week). The timeline is a slide source, not something to paste into the work report.
- Keep text on the slides minimal — usually no more than a short headline.
- Put the narrative in the speaker notes. Write complete speaker notes for every slide explaining what I should say, timed so that the entire presentation fits within 3 minutes.
## Catching Up Mid-Week
If these instructions land mid-week, we may have missed deadlines or be missing files you would usually create earlier in the week. That's okay. Just get me up to date, create any missing files, and email me the drafts of the Work Report and slideshow. If the automated deadline to send the Work Report has passed, send it as soon as it's ready instead.
Open prompt → I want to improve how my Fractal Accelerator homework process is working for me. You've been helping me run it; now help me tune it.
First, find my Fractal Accelerator project folder and read `AGENTS.md`, `homework.md`, `goals.md`, `tasks.md`, and `devlog.md`.
Then review how the last week actually went, and tell me plainly:
1. Where the process worked — what got shipped, reported, or demoed because the system did its job.
2. Where it broke down — deadlines that surprised me, files that went stale, steps I skipped, moments where you were too verbose, too proactive, or asked too much of me.
3. Where I seemed confused — anything I asked about repeatedly or got wrong (like not realizing the weekly goal is a week-long project).
Then ask me, one at a time, at most three questions about what I want to change — for example: how much you should write, how you should speak to me, how often you should check in, how much you should do without asking.
Finally, apply the changes: update `homework.md` and `AGENTS.md` so the process matches what we agreed, under this week's dated heading where appropriate. Keep the core structure intact — daily update, Thursday drafts, Friday work report, Saturday demo — unless I explicitly say otherwise. End by summarizing in five sentences or fewer exactly what will be different starting tomorrow.
Open prompt → I'm a student in the Fractal Accelerator. It's time to set my goals for this week. Act as my collaborator for a short, focused goal-setting session — this should take about 15 minutes, not an hour.
First, find my Fractal Accelerator project folder and read `goals.md`, `tasks.md`, and `devlog.md` if they exist.
Then run the session:
1. **Look back, briefly.** In a few sentences, tell me what shipped last week and what stalled, based on the files. If this is week one or the files are empty, skip this.
2. **Ask me one question:** what do I want built and working by Friday? This is my top priority for the week — one thing, not five.
3. **Sharpen it with me.** If my answer is vague, push until it's concrete enough that on Friday we could both point at it and say "done" or "not done." If it's too big, help me cut scope until it fits in a week. It should be demoable in 3 minutes on Saturday morning.
4. **Break it down.** Turn the goal into concrete tasks spread across the week — this is a week-long project, not a one-sitting sprint. Front-load the scary or uncertain parts.
5. **Write it down.** Record the goal in `goals.md` and the tasks in `tasks.md` under a dated heading for this week (e.g. `## Week of 2026-07-20`), keeping the current week on top.
Keep the whole session tight. Don't ask more than one question at a time, don't ask about my schedule or hours, and don't pad your responses. When we're done, give me the goal in one sentence and the single best first task to start on today.
Open prompt → My daily briefing email isn't working well for me yet. I want you to diagnose it and fix it.
First, find my Fractal Accelerator project folder and the daily automation that sends my morning update. Read the automation's instructions, `homework.md`, and the last several daily updates you sent me (check your own records or sent mail).
Then diagnose honestly. Compare the recent updates against my actual work — `tasks.md`, `devlog.md`, and files that changed on those days — and tell me:
1. Are the updates tracking my real work, or are they generic and repetitive? Quote an example if they're same-y.
2. Are they too long, too verbose, or padded with filler?
3. Is the voice right, or does it sound like a corporate bot?
4. Is anything useful missing — deadlines I lose track of, blockers you never surface, decisions you never ask me for?
Then ask me two quick questions: what do I actually want from this email each morning, and how do I want it to sound?
Finally, rewrite the daily-update instructions — in `homework.md` and in the automation itself — so tomorrow's email is short, in my voice, and grounded in what I actually did the day before. Show me a sample of what tomorrow's email will look like before you finish.