# WORK ORDER — Station 3: first contact (GLAMMBOX pack)

Teach a new user the single most important boundary before they touch an
agent: a chat model sees only what is typed or uploaded into it, nothing on
the computer, until a folder is explicitly opened to it.

## Goal
The person can state, correctly, what the model could see before and after
opening one folder to it, and that folder exists.

## Prerequisites
- The app from Station 2, signed in.
- One PC with a normal file system (Windows, Mac, or Linux all work).

## Steps
1. In the chat app, ask directly: "What can you see about me and this
   computer right now?" Read the answer together. A plain chat window can
   only see the words typed into it and any file explicitly uploaded to
   that conversation — nothing else, not the desktop, not other apps, not
   other browser tabs.
2. Confirm the boundary out loud: no folder access, no file list, no
   screen, no other app, unless the person deliberately connects one.
3. Create one dedicated folder on the PC for this work, for example
   Documents/AI-Home. This is the "home folder" — the one place an agent
   will later be allowed to read and write.
4. Put one real, non-sensitive file inside it (a to-do list, a project
   note — nothing containing passwords, banking details, or medical
   records).
5. State the rule out loud, and mean it: one folder, on purpose, never the
   whole Documents drive, never Downloads, never the whole home directory
   "to be safe." An agent that can touch everything is not safer — it is
   just less supervised.

## Verification
- The home folder exists and has at least one real file in it.
- The person can explain, unprompted, the difference between "the chat
  window" (sees only the conversation) and "an agent with a folder" (sees
  what that folder contains, and nothing outside it).

## Honest costs
Free. This station costs nothing but five minutes and one new folder.