Preserve scheduled-job outcomes
Capture stdout, stderr and exit status separately, publishing a report only after a successful job.
Three outcomes travel together
A scheduled command produces standard output, standard error and an exit status. They answer different questions. Standard output may be a report; standard error carries diagnostics; the exit status says whether the command completed according to its contract. A nonempty output file alone does not prove success.
Run a trusted job with its streams redirected to separate files. Use an if/else form that preserves the original status in the else branch. The pattern if ! command negates the status, so reading $? inside that branch would observe the negation rather than the failed command’s code.
Do not publish partial success
Write the job’s report output to a temporary file. On success, move that complete file into the published path. On failure, retain the previous report, keep diagnostics in the log, print a concise failure message to stderr and return the job’s nonzero status.
An EXIT trap cleans the temporary directory on either path. This is a small example of reliable automation under a scheduler, where nobody is watching the terminal. It does not install cron or a system service in the browser. The checker substitutes failing jobs that print partial output, proving that a plausible-looking file cannot bypass the success requirement.
if bash ./jobs/report.sh > example-output.txt 2> example-errors.txt; then
printf 'Example job succeeded\n'
else
status=$?;printf 'Example job returned %d\n' "$status"
fi
Put it into practice.
Run the trusted jobs/report.sh in a child Bash. Capture its stderr exactly in job.log and its stdout in a temporary file. On exit 0, replace report.txt with the complete stdout and print Report published plus a newline. On nonzero exit, leave any prior report.txt unchanged, print Job failed: <status> plus a newline on stderr, and exit with the original status. Do not publish partial stdout or duplicate job diagnostics on your own stderr.
See an explained solution
tmp=$(mktemp -d "${TMPDIR:-/tmp}/rocks-job.XXXXXX") || exit 1
trap 'rm -rf "$tmp"' EXIT
if bash ./jobs/report.sh > "$tmp/report.txt" 2> job.log; then
mv "$tmp/report.txt" report.txt || exit 1
printf 'Report published\n'
else
status=$?
printf 'Job failed: %d\n' "$status" >&2
exit "$status"
fi
The success branch publishes a complete report, while the failure branch preserves the previous result and the original exit code. Diagnostics remain available in job.log without being mixed into structured report data.