How to set up a first session, inspect the interface, and understand the approval model.
The first hour is intentionally Claude-specific. The Claude Code interface controls what the tool can see, what it can change, and how the session is organized. Start there. Once you are comfortable with it, the methods extend to other contexts.
What You Need
A paid Claude plan
Pro, Max, Teams, or Enterprise. The free tier doesn't include Claude Code.
A spreadsheet, a set of documents, a project. Anything you want Claude to work with.
That is enough for a first session, provided you know the folder well enough to judge the results.
The Claude Desktop Interface
Claude Desktop has three tabs. Chat is regular conversation, like using Claude on the web, with no file access. Cowork runs tasks in the background on Anthropic's servers. The one that matters here is Code, which can read your files, edit them, and run commands on your computer. If your files are in a folder and you want Claude to work with them directly, the Code tab is where you start.
Why it's called "Code"
The name refers to the tab's capabilities (file access, editing, running commands), not to programming. You can use it with spreadsheets, metadata records, documents, or any other files in a folder.
Claude Desktop App
Chat
Code
Cowork
EnvironmentLocal (your computer)
FolderSelect folder... ← pick your project here
ModelSonnet (default, fast and capable)
PermissionsAsk permissions (default, safest)
Type your message here...
Setting Up Your First Session
Click the Code tab at the top of the app. If it asks you to upgrade your plan or sign in, follow those steps first.
Set Environment to "Local", which means Claude will work with files on your own computer.
Click "Select folder" and choose the folder you want Claude to work with. This could be a project folder, a folder of spreadsheets, a set of documents. Whatever you want help with.
Leave the other settings at their defaults. "Ask permissions" means Claude will ask before changing anything, and the default model (Sonnet) handles the tasks in this guide well.
Suitable first folders
A few spreadsheets, a set of metadata records, or an inherited project directory. Pick a folder where you would notice if the output were wrong.
Begin with Inspection
Ask Claude what is in the folder before you ask it to change anything. Once you and the tool agree about what is there, your follow-up prompts can be more specific, and you will know what a correct answer looks like.
You type:
What's in this folder? Give me an overview of the files and what they contain.
Claude reads through the files and reports back: file types, sizes, and what it can infer about the contents. None of this changes anything on disk, so you can ask as many exploratory questions as you like before deciding what to do next.
You type:
Open the spreadsheet called survey_responses.csv and tell me about the columns. Are there any obvious data quality issues?
You type:
I have 200 records in the records/ folder. Can you check whether any are missing required fields?
You type:
Look at every file in this folder and tell me which ones mention "digital preservation."
None of these prompts trigger an approval step. They only read files, so there is nothing to accept or reject.
When Claude Wants to Change Something
When you ask Claude to edit a file, create something new, or run a command, it pauses and shows you what it plans to do before anything happens. Here is what that looks like:
You type:
That CSV has some duplicate rows. Remove the duplicates and save the file.
Claude will show you a diff view (a before-and-after comparison of what will change in the file). Removed lines appear in red, added lines in green. Check that the changed lines correspond to what you actually asked for, and then decide:
Accept: the change goes through
Reject: nothing changes, and Claude asks how you'd like to proceed differently
Claude Code uses Git (a version-control system) to track every change, so if you accept an edit and later decide it was wrong, you can roll it back. Git in Practice explains how.
A Real Example: Cleaning a Spreadsheet
A routine cleanup task makes a good first example. You already know what the cleaned file should look like, so you can judge each step as it happens.
You
Look at ill_requests_2025.csv and tell me what shape it's in.
Claude
The file has 847 rows and 12 columns. The "date_requested" column has three different date formats (MM/DD/YYYY, YYYY-MM-DD, and Month DD, YYYY). 23 rows have a blank "patron_type" field. And 14 rows are exact duplicates.
You
Standardize the dates to YYYY-MM-DD, remove the duplicate rows, and add "REVIEW" in the patron_type column wherever the value is missing.
Claude
Shows diff: 14 duplicate rows removed, 23 blank patron_type fields filled with "REVIEW", all dates converted to YYYY-MM-DD format.
Claude wants to edit ill_requests_2025.csv: Accept / Reject
You
Accept.
Claude
Done. The file now has 833 rows. All dates are in YYYY-MM-DD format, and 23 rows are marked "REVIEW" in the patron_type column so you can fill those in.
You
Now create a summary. How many requests per month, broken down by patron type?
Things Worth Knowing Early
You can interrupt Claude at any point if the work is going in the wrong direction. Be specific even early on. State the task as a concrete outcome. If you want Claude to look at a particular file, drag it into the window or type @ and the filename.
Review diffs before accepting them, and spot-check the output. Common early mistakes are easy to miss. A misread column, a default format you never asked for, a transformation applied too broadly. Each session starts fresh, with no memory of earlier ones. If you find yourself repeating the same instructions across sessions, write them into CLAUDE.md, which Claude reads at the start of every session.
Cost awareness
Claude Code usage counts against your plan's limits. Reading files uses relatively little; larger jobs use more. Type /cost in the chat to check your session usage.
Things to Try in Your First Session
Explore unfamiliar files
Point Claude at a folder you inherited or haven't looked at in a while.
"What's in this folder and how is it organized?"
Check a spreadsheet
Have Claude audit a data file for quality issues.
"Look at this CSV. Are there missing values, inconsistencies, or obvious errors?"
Search across documents
"Search every file in this folder for mentions of 'accessibility' and summarize what you find."
Reformat something
"Convert this tab-separated file to CSV and standardize the date column to YYYY-MM-DD."
All of these produce output you can verify against your own knowledge of the material.
Where to Go from Here
From here, the next chapters cover prompting technique and more specialized workflows.