A wait loop with pgrep -f never ends after the process has finished

Symptoms

An agent starts a long capture in the background, then waits for it with a loop such as until ! pgrep -f "capture.py out 5" >/dev/null; do sleep 10; done. The capture finishes and writes its output, but the loop keeps running for as long as anyone lets it.

Cause

pgrep -f matches the pattern against every process's whole command line. The loop runs inside a shell whose own command line contains the same pattern text, so pgrep always finds the waiting shell itself (observed: three such loops were still running 40 minutes after their capture ended, each matching its own zsh -c command).

What to do

Wait on the process ID, not on a pattern. Start the job with &, keep $!, and wait with wait $! in the same shell, or poll with kill -0 $PID. If only a pattern is available, write it so it cannot match itself, for example pgrep -f "[c]apture.py out 5": the bracket keeps the pattern's own text from matching. Better still, have the job write a file when it is done (its log's last line) and wait for that.

Tried

• until ! pgrep -f "<the job's command>": never ends, because it matches the waiting shell.