Start here · 30 minutes

Start Here

Install Claude Code and Set Up Your Research Desk

You'll have
Claude Code running in the desktop app, a dedicated Investing folder with the course's starter files, and a first answer about a company you know, with its sources and a note of what Claude could not verify.
You need
A Mac or Windows computer and a paid Claude plan that includes Claude Code. No coding, no terminal.
Video

The 12-minute video for this lesson is on its way. Meanwhile, the Guide me prompt takes you through the same steps.

In one minute

A chatbot in a browser answers from memory. Claude Code works in a folder on your computer: it can fetch a filing, read every page, save a table and come back to it next week, and it asks before it does anything you have not allowed. In this lesson you install it (about fifteen minutes), point it at a quiet new folder called Investing, set the safety switch to the right position, and build the research desk the other lessons use: a brief for Claude, a verification checklist, a watchlist and two folders. Then you ask a first real question about a company and check the answer yourself.

Install first (Steps 1 to 3 below). Then paste the Guide me prompt, or take the Quick win.

Guide me

Once the Code tab is open on your new folder (Steps 1 to 3 below), paste this and reply "next" whenever you are ready to move on.

Guide me prompt
You are my tutor for Lesson 0 of the Latticework AI Bootcamp 2.0, "Start Here: Install Claude Code and Set Up Your Research Desk". I am an investor, not a programmer, and new to Claude Code. By the end I want a safe setup, the course's starter files in this folder, and one sourced answer about a company I know.

How to teach me:
- One step at a time. At the end of every step, STOP and ask me to reply "next" (or to ask a question). Never start the next step before I reply.
- Plain English, short. Define any technical term in one line the first time you use it.
- Before you create or change a file, or run a command, tell me in one sentence what it will do and why.
- Work only inside this folder. Never install anything without asking me first and explaining why.
- Keys and passwords go only in a file called .env, typed by me in a text editor; never ask me to paste one here, and never show me the contents of .env. My name and email for the SEC I may type here; save them to .env as SEC_USER_AGENT.
- If something fails, tell me in plain words what happened, then fix it.
- Every fact you give me comes with its source and a link. If you cannot verify something, say "I could not verify this" instead of guessing.
- You cannot see my screen. When you need to know what the app shows, ask me to read it to you.

Step 0 - Catch-up check. Tell me the full path of this folder. If it is my home folder, Desktop, Downloads or full of unrelated files, stop: tell me how to make a new empty folder (for example Documents/Investing) and open it with the folder control under the prompt box. Otherwise look for CLAUDE.md, always-verify.md, watchlist.md, notes/, data/, .env and bootcamp-progress.md, and tell me in three lines what you found. Create nothing yet.

Step 1 - Permission modes. Ask me which mode the selector under the prompt box shows. Explain in one line each: Manual, Accept edits, Plan, Auto, Bypass permissions. Recommend Manual today, Plan before anything bigger, never Bypass permissions. Tell me how to change mode (Ctrl+Shift+M, or Cmd+Shift+M on a Mac) and how to stop you (the stop button, or Esc).

Step 2 - The research desk. Show me the plan and wait for my OK. Then create: CLAUDE.md (a short starter brief: I am a long-term investor; always cite sources with form, date and link; never invent numbers, quotes or dates; say "insufficient evidence" when the evidence is thin; ask before installing anything), always-verify.md (a starter checklist: every number, every quote, every date, every currency and unit, every ownership claim, anything surprising), watchlist.md (two headings: ## Companies, one ticker per line, ask me for up to ten, empty is fine; and ## Phrases, empty for now), empty folders notes/ and data/, and bootcamp-progress.md. Before the first file, tell me what approval request I will see and what to read in it. Then explain in five lines what each file is for.

Step 3 - Python check. Explain in two lines why Python matters (later lessons have you write small programs that download and tabulate data) and that the desktop app does not include it. Check whether Python 3 is installed. If not, explain my options (Windows: python.org, the Microsoft Store, or winget; Mac: the official python.org installer), recommend one, and install nothing unless I say yes; then check again.

Step 4 - A first useful answer. Ask me for a US-listed company I know well, and for a name and email for the SEC (its rule for automated visitors); save them in .env as SEC_USER_AGENT and explain .env in one line. Then, from SEC EDGAR, tell me in under 200 words what the company does and how it makes money (quoting its latest 10-K), when that 10-K was filed and which fiscal year it covers, with links. End with "What I could not verify". Save it to notes/first-answer-<TICKER>-<date>.md. Ask me to open the 10-K and check the filing date and one quoted phrase myself (Ctrl+F, or Cmd+F on a Mac), and wait for my result.

Step 5 - Habits and limits, in under 120 words: a new session (+ New, or Ctrl+N / Cmd+N) rereads CLAUDE.md; "remember X" goes to your auto memory, while "add this to CLAUDE.md" changes my brief; the usage ring under the prompt box and the five-hour and weekly limits, shared with Claude in the browser; the three habits (this folder only, keys only in .env, read the plan before approving). Then ask me for one preference about your answers, add it to CLAUDE.md, and show me the change.

Final step - Wrap up. List every file we made, with its path. Add a dated line to bootcamp-progress.md: "Lesson 0 done: research desk set up; Python <version or not installed>; first answer on <TICKER>." Suggest a next lesson from what I said: Lesson 1 (courses.latticework.com/bootcamp/verification) if I distrust AI answers, Lesson 2 (/bootcamp/analyst-brief) if I want answers in my own style, Lesson 3 (/bootcamp/sec-filings) if I want the most useful thing first.

Start with Step 0 now.

Quick win

Already have Claude Code open in a new, empty folder? This sets up the research desk the course uses.

Quick win prompt
Set up this folder as my investment research desk. I am an investor, not a programmer. Before you create anything, show me the plan and wait for my OK. Create: CLAUDE.md (a short brief for you: I am a long-term investor; always cite sources with form, date and link; never invent numbers, quotes or dates; say "insufficient evidence" when the evidence is thin; ask before installing anything), always-verify.md (a starter checklist: every number, every quote, every date, every currency and unit, every ownership claim, anything surprising), watchlist.md (two headings: ## Companies, one ticker per line, ask me for up to ten; and ## Phrases, empty for now), bootcamp-progress.md (one dated line: "Research desk set up"), and empty folders notes/ and data/. Then explain in five lines what each file is for and how I can change it later.

Then ask your first real question. Replace the ticker with a company you know well.

Quick win prompt
Ticker: COST. Using SEC EDGAR (ask me for a name and email for the SEC User-Agent before your first request, and ask before fetching anything), tell me in under 200 words: what this company does and how it makes money, quoting its latest 10-K; the date that 10-K was filed and the fiscal year it covers; and how many business segments it reports. Give a link to every source and, for each quote, a short phrase I can search for in the filing to check it. End with a section headed "What I could not verify". Save the answer to notes/first-answer-<TICKER>-<date>.md.

Why this matters for investors

Most investors who try AI stop at the chat window. You ask, it answers fluently, and you are left wondering whether the answer came from last quarter's filing or from something the model half-remembers from two years ago. That doubt is healthy, and it is also expensive: if every answer has to be rechecked from scratch, the tool saves little.

Claude Code changes the arrangement. Because it works in a folder, it can go and get the document, save it, and quote from the copy you can open yourself. Because the folder persists, the work accumulates: a brief that tells Claude how you invest, a checklist of what you always verify, a watchlist, and dated notes you can return to.

The set-up is also where safety is decided. An assistant that can create files and run programs deserves the same care you would give a new analyst on day one: its own desk, clear rules, and a habit of showing you its plan before it acts. Thirty minutes here make every later lesson faster and calmer.

The idea

Claude Code is Claude with a desk and a pair of hands. In a browser, Claude can only talk. In Claude Code it can also read and write files in one folder, fetch web pages, and run small programs, all on your computer. You stay in charge through three controls.

The folder. Claude works inside the folder you open. Make it a quiet, new folder used only for research, never your Desktop, Downloads or home folder, where it could read files that have nothing to do with investing. The course uses the same layout everywhere:

In your folder What it is
CLAUDE.md The standing brief Claude reads at the start of every session here
always-verify.md Your checklist of what gets checked before it reaches a decision
watchlist.md The tickers you follow
notes/ Everything Claude writes for you, by date
data/ Filings and data downloads
.env The private drawer for keys and your SEC name and email; never shared
bootcamp-progress.md One line per lesson: what it set up

The permission mode. It decides how often Claude asks before acting. Manual asks before every file change and every command, and shows you exactly what will change. Accept edits makes file changes without asking but still asks before other commands. Plan explores and proposes a plan without changing anything. Auto, if your app offers it, runs without routine prompts while a background check reviews each action. Bypass permissions removes the checks entirely; it is hidden unless you switch it on, and the course never uses it.

Your attention. Read the plan before you approve it. That one habit catches most mistakes before they happen.

A concrete example. Ask a browser chatbot, "When did Costco file its latest 10-K?" and you get a date from memory, possibly a year stale, stated with full confidence. Ask Claude Code the same thing in your folder and it looks up the filing list on SEC EDGAR, gives you the date with a link, saves the answer to notes/, and tells you what it could not confirm. You open the link and see the same date. That is the whole course in one exchange: the answer arrives with its address, and you check it.

Step by step

Step 1: Choose a plan and install the app

Claude Code needs a paid Claude plan; the free plan does not include it. At the time of writing, Pro costs $20 a month, or $17 a month billed annually; Max costs $100 a month and Ultra $200 a month, with more usage. Team and Enterprise plans include it too. Check claude.com/pricing for current terms. Pro is enough to start.

A word on usage limits. Paid plans meter usage in two ways: a five-hour session window, which starts with your first message, and a weekly limit. Both are shared between Claude in the browser and Claude Code. Ultra has no five-hour limit on verified accounts. The usage ring under the prompt box shows where you stand. If you hit a limit, you can wait for the reset, upgrade, or turn on paid usage credits. Long tasks with many parallel helpers use the allowance fastest.

Then download the Claude desktop app for Mac or Windows from claude.com/download, or follow Anthropic's desktop quickstart. Run the installer. The app includes Claude Code; you do not need to install anything else for it.

Step 2: Sign in and open the Code tab

Launch Claude (Applications on a Mac, the Start menu on Windows) and sign in with the account that holds your plan. Click Code at the top of the sidebar. If the tabs show only icons, hover over them to see the names.

  • A prompt to upgrade means the account has no plan that includes Claude Code.
  • A prompt to sign in online: complete it, then restart the app.
  • A "403" error: sign out and back in.

Prefer the terminal? Claude Code also runs from the command line; Anthropic's quickstart has the one-line install. Almost everything in this course works there too; scheduled tasks (Lesson 9) and the Browser pane (Lesson 10) need the desktop app.

Step 3: Make a quiet folder and open it

Create a new, empty folder, for example Investing inside Documents. Under the prompt box is a row of controls. From the left: the environment selector (choose Local, so Claude works on your own computer), the folder control (choose Open folder… and pick Investing), the mode selector and the model picker. Leave the model on its default.

If the environment and folder controls sit above the prompt box instead, you have an older build: update it with Help > Check for Updates… on Windows or Claude > Check for Updates… on a Mac.

Step 4: Set the permission mode, and learn the brakes

Look at the mode selector before your first message; the app remembers your choice for each folder. Pick Manual for now. Open the menu with the selector or with Ctrl+Shift+M (Windows) / Cmd+Shift+M (Mac). Older videos and blog posts use earlier names: "Ask permissions" is now Manual and "Auto accept edits" is now Accept edits.

  • How changes are shown. In Manual mode, each file change appears as a diff, the lines being added and removed, with buttons to approve or decline. In Accept edits, a small indicator such as +12 -1 appears; click it to review.
  • How to stop. Click the stop button or press Esc. To change course without stopping, type a correction and press Enter; Claude reads it after its current action.
  • Plan first. For anything bigger than a small task, switch to Plan: Claude explores, then proposes, and changes nothing until you approve.
  • A new session. + New in the sidebar, or Ctrl+N / Cmd+N. Each session starts fresh but reads CLAUDE.md again, so your brief carries over.

Step 5: Build the research desk

Paste the first Quick win prompt above, or the Guide me prompt, which does the same with explanations. Claude shows its plan; read it, then approve. In Manual mode you will approve each file. Open the folder in Explorer or Finder and look: the files are ordinary text you can open in any editor.

Step 6: Check for Python

Many lessons have Claude write a small program in Python, a widely used programming language, to download filings or tabulate data. You never write it; Claude does, and explains it if you ask. The desktop app does not include Python, so check:

Step 6: Check for Python
Check whether Python 3 is installed on this computer and tell me the version. If it is not installed, explain my options in plain English, recommend one, and wait for my yes before installing anything. Explain each step before you run it.

On Windows the usual routes are the installer from python.org, the Microsoft Store, or winget, the package manager built into Windows. On a Mac, use the official installer from python.org. If you would rather not install anything today, say so; Claude can often work around it.

Step 7: Ask a first question, and check the answer

Paste the second Quick win prompt with a company you know. When the answer arrives, open the 10-K link in your browser, press Ctrl+F (Cmd+F on a Mac) and search for one of the quoted phrases. Check the filing date at the top of the filing page. Read the "What I could not verify" section: an assistant that tells you where its knowledge stops is more useful than one that never does.

Claude will ask before it fetches a web page. The prompt offers an option to stop asking for that site; for the SEC's sites (www.sec.gov, data.sec.gov, efts.sec.gov) that is reasonable. Use that option only for a website, never for a general command such as python or curl.

Make it yours

  • Edit the starter brief now. Open CLAUDE.md and add one line about how you invest. Or ask Claude, "add this to CLAUDE.md: …". (Saying "remember …" puts the note in Claude's own auto memory instead.) Lesson 2 writes the full brief from an interview.
  • Fill the watchlist. Ten tickers you follow closely. Later lessons suggest companies from it.
  • Pin your safety rule. If you work with confidential client material, add a line to CLAUDE.md such as "Never upload or send anything from this folder to a website."

Checkpoint

Common pitfall

Opening Claude Code on the wrong folder. If you point it at your home folder or Desktop, everything there is within reach: tax returns, client files, downloads you forgot about. Nothing bad has to happen for this to be a mistake. Give Claude its own desk, and keep private documents off it unless a task needs them.

A second one: switching the brakes off to go faster. Auto and Bypass permissions are tempting once the approval requests start to feel repetitive. Use Accept edits for routine work, Plan for anything new, and leave Bypass permissions off; it is meant for isolated test machines, not a computer with your files on it.

Go deeper (optional)

Where to next

  • Lesson 1, Fluent vs. Right. Why AI answers sound right when they are wrong, and your personal checklist.
  • Lesson 2, Your Analyst's Brief. Turn the starter CLAUDE.md into a brief that sounds like you.
  • Lesson 3, The Primary Source. The most useful thing Claude Code does for investors: read and compare 10-Ks.

Or open the Prompt Library for every prompt in the course on one page.

All lessons