Recorded shell

Work in your usual terminal. Every command you type is recorded, with nothing to add in front of it.

kveritas shell
(kv) $ python train.py
(kv) $ ./bench --size 1024
(kv) $ exit

The first time, it asks what the report should show (results only, file names, or your code too) and how to get your results.

What gets recorded

For each command:

• time, CPU, memory, GPU and energy use,
• your results, from KV lines or autolog,
• what it printed,
• whether it worked or failed. Failed commands are kept on record too.

Simple commands like cd, export, clear, and editors such as vim, are not recorded.

When you exit

1. You see the latest value of each result.
2. Type the names of the ones you want to claim.
3. Choose whether to seal the report now.

Not sealed yet? Run kveritas shell again later and you pick up where you left off.

Can it be fooled?

• Your programs can't pretend a command started or ended. Only the shell can.
• Recording starts before each command does.
• Each command is reported to the server as soon as it ends, so it can't be removed later.
• Only one recorded shell per project at a time.

Works with bash 4.4 or newer and zsh, on Linux and macOS.