07 · Run your first task

Client installed, signed in, window understood. Time to run something real.

Three tasks, increasing in difficulty, each done in ten minutes. Pick the one closest to your actual work — running one beats reading three chapters.

Before you start

1. Make an empty folder for the workspace, e.g. wb-trial on your desktop.
2. Copy the files you want processed into it. Do not use the originals.
3. Confirm you are still on default permissions.

Task one: turn raw material into a structured document

The easiest win, five minutes to a result.

Preparation

Drop a piece of raw material into the workspace — a meeting transcript, scattered notes, a long email, a client brief. Plain text, Word, PDF and Markdown all read fine.

The prompt

Read meeting-0312.txt in the workspace and turn it into formal minutes containing:

1. Meeting subject and time
2. Key conclusions, one sentence each
3. An action-item table: item / owner / due date
4. Disputed or unresolved items, listed separately

Output as a Markdown file saved to the workspace, named minutes-0312.md.
No preamble and no closing summary boilerplate.

Swap in your own filename.

Why it is written that way

• An exact filename saves a round of it guessing which file you meant
• Four named sections mean it does not invent a structure, and you know what to check
• A table for action items is far more precise than "organise it"
• An explicit output format and filename — without them you get a block of chat text that never lands on disk
• The last line kills "Here are the minutes I have prepared for you…", which is the default habit

Acceptance

• [ ] minutes-0312.md appears under Artifacts on the right
• [ ] All four sections are there
• [ ] The action items really are a table, and no owner names were invented
• [ ] The conclusions exist in the source material

Check for invented content first

This is the first acceptance test for any AI output. Watch the owner and due date fields especially — if the source did not say, the correct behaviour is to leave it blank or write "not specified", not to fill in something plausible.

If it invented something, reply: For action items with no stated owner, write "not specified" — do not infer.

Task two: make a spreadsheet explain itself

This one produces charts you can drop straight into a report.

Preparation

Put an Excel or CSV file with real data in the workspace. Sales records, project hours, inventory, survey results — anywhere from dozens to thousands of rows.

The prompt

Read sales-2026H1.xlsx in the workspace and tell me only:

1. What fields it has, and what you understand each one to mean
2. How many rows, and whether there are missing or clearly anomalous values

Answer only those two points. Do not start analysing.

Run that first. Confirm it read the fields correctly, then send the second round:

Correct. Now:

Aggregate revenue and margin by month and product line. Produce a bar chart
(revenue) and a line chart (margin), plus a conclusion of under 200 words
naming the two changes most worth attention.

Output as an HTML page with interactive charts, saved to the workspace.

Why two rounds

Because a spreadsheet is the input most easily misread. Abbreviated field names, merged cells, multiple header rows, hidden columns — any one of them produces a result that looks entirely presentable and is entirely wrong.

Thirty seconds confirming it read correctly beats three rounds of rework. This is the "small steps" principle from 08 Briefing it properly in its most useful form.

Acceptance

• [ ] The field meanings it listed match your understanding
• [ ] The chart figures reconcile with the source (spot-check two or three months by hand)
• [ ] The conclusion points at changes that actually exist in the data
• [ ] The HTML opens in the Browser view on the right and the charts respond to hover

Task three: tidy a messy folder

The first task that really changes your files. Do it on a copy.

Preparation

Copy a chaotic folder into the workspace — a downloads directory, a pile of receipts, a batch of photos. Copy, not the original.

The prompt

Read every file in this folder and rename them in bulk according to
"date + subject + type", derived from file content or timestamps.

Show me a preview first and execute only after I confirm.

Present the preview as a table: original name → new name → basis.
Do not modify any file yet.

You get a comparison table.

Read every row. Look especially for:

• Two different files ending up with the same name
• Whether the date came from the content or was guessed from file properties
• Any file that got skipped

Once satisfied:

The table looks right, execute the rename. When done, tell me how many
were renamed, how many were skipped, and why.

Acceptance

• [ ] Nothing was modified during the preview stage
• [ ] The file count after execution matches the count before (nothing lost)
• [ ] It reported skipped files and reasons
• [ ] The Changes view on the right matches the comparison table

The red lines for this task

1. Copies only. Bulk renaming is irreversible. The system does auto-backup, but do not treat that as your only insurance.
2. Do not run this over folders holding private data, financial credentials or keys. Pull the files you want processed into a separate temporary directory first.
3. System directories, encrypted files and files locked by other programs are skipped automatically. That is protection working, not a failure.

What the three tasks proved

Task	Capability tested	Habit learned
One · material → document	Reading files, structured output, writing to disk	Specify structure and filename, cut the boilerplate
Two · spreadsheet → charts	Data comprehension, aggregation, visualisation	Confirm it read correctly before letting it calculate
Three · bulk rename	Real local file operations	Preview, then execute, checking row by row

With those three done you have used WorkBuddy's three core capability paths. Everything else builds on them.

When it does not work

Symptom	Try this first
Cannot read PDF / images / Excel	Switch to a model that takes image input, or install a document Skill
The generated file will not open	Next time name the format explicitly: Word / Excel / PDF / Markdown
Garbled or nonsensical output	Switch model and retry; split the task smaller
Stuck and unresponsive	Stop, switch model, run again
Consistently slow	Check the network; then check whether you handed it one huge compound task

Full troubleshooting in Troubleshooting.

Next

It runs, but the results are not good enough? That is a prompt problem, not a product problem: 08 Briefing it properly.