Files, folders, paths

What you will learn How to read a path, what a file extension actually promises, and the split between files people can read and files only machines can.

A path is an address

Folders sit inside folders. Written on one line, that nesting is a path.

text

macOS · Linux    /Users/jinho/work/report.mdWindows          C:\Users\jinho\work\report.md

Piece	Meaning
`/` or `C:\`	the outermost point — start here
`Users/jinho`	your account's folder
`work`	a folder inside it
`report.md`	the file name and extension

The separator differs by operating system. macOS and Linux use /, Windows uses \. An example starting with /Users/… is a macOS one.

Absolute and relative

	Example	Meaning
Absolute	`/Users/jinho/work/report.md`	the same place, called from anywhere
Relative	`work/report.md` · `./report.md`	relative to the current folder
Up one	`../report.md`	the folder above

Give agents absolute paths. A relative path depends on "where you are right now", and that "now" can differ between you and the agent.

Registering a working folder is precisely the act of pinning that down.

An extension is only a promise

The .md, .csv, .pdf at the end is the extension. It does not change the contents — it is a label saying which program should open it.

text

✗ renaming report.csv to report.xlsx makes it an Excel file✓ only the label changes — the contents are still CSV, so it opens broken

Windows hides extensions by default. What looks like report.pdf may really be report.pdf.exe. Turn extensions on in File Explorer — it is basic security hygiene.

Files people read, files only machines read

This split keeps coming back later.

	Text files	Binary files
Opened raw	readable	garbage characters
Examples	`.txt` `.md` `.csv` `.json` `.py`	`.pdf` `.xlsx` `.hwp` `.png` `.zip`
Change history	visible line by line	a wholly different file
For AI	feed it directly	needs an extraction step

That last row is the practical one. It is why "conversion" appears whenever an agent handles a PDF or a Hangul document — picked up in Turning documents into material and Shapes of data.

Naming — it saves you later

Agents produce files; people look for them later. Names bridge the two.

text

✗ final.md · final_really_final.md · report (1).md✓ 2026-08-19-weekly-report.md · q3-returns-analysis.csv

text

□ Hyphens over spaces — spaces need quoting on the command line□ Dates as 2026-08-19 — alphabetical order becomes chronological order□ Non-ASCII names work, but can break when uploaded to other systems

Hidden files

A name starting with a dot (.env, .gitignore) is not shown by default. Configuration files use this convention, so half of "there is no such file" is a file that is simply hidden.

.env deserves special care. It holds passwords and keys, it is hidden by convention, and sharing it by accident is an incident on its own — see Accounts, passwords, API keys, permissions.

Common misconceptions

"Why can't I give a large file to the AI?"

It is about characters, not megabytes. A model sees a limited amount at once (the context window), so long documents are split and only the relevant part is passed. A 10MB photo costs far less than 1MB of text.

"I moved the folder and the program cannot find it"

Because a path is an address. A registered absolute path does not follow the folder. If you move it, update the registration.

Check yourself

1. When is an absolute path safer than a relative one?

Answer

When "the current folder" may differ between parties — telling an agent where a file is, for example.

2. Does renaming .csv to .xlsx produce an Excel file?

Answer

No. The extension only says which program should open it. The contents stay CSV and the file opens broken.

3. Which is easier to hand straight to an AI — text or binary?

Answer

Text. Binaries (PDF, Excel, Hangul documents) need a conversion step to pull the characters out first.

Next, the window where you handle those files directly → The terminal and commands