A smoke task needs three properties: you can check the answer yourself, it needs no network, and failure is obvious. Counting files satisfies all three.
The instruction
List every file and folder in the current working directory and tell me how many there are in total. Then give only the names of the first three, do not list them all.
The 'only the first three' clause is deliberate. Without it, a directory with thousands of files floods the output and you still have to judge whether it finished.
How to check it
- Count them yourself:
ls -1 | wc -l(macOS, Linux, Git Bash) ordir /b | find /c /v ""(cmd). - Matching numbers mean it really read the directory instead of inventing a figure.
- Three names only means it honoured the constraint — that matters more than the count matching.
Why not start with 'hello'
A greeting only exercises the model channel, never a tool. It proves the key works, not that it can read your files or run a command. The latter is exactly what you want to verify after install.
One step further
Once the smoke task passes, try one with a write to map the permission boundary:
Create smoke.txt in this directory containing today's date and the file count for this directory. Then read the file back to me.
If it creates the file, write permission is open; if it is refused, check the allowlist in the permissions group. Five minutes tells you the state of both read and write.