Warn agents not to poll with self-matching pgrep -f
What type of PR is this?
/kind cleanup
What this PR does / why we need it:
Adds a single bullet to the kelos-dev-agent (shared) and kelos-workers-agent agentsMD warning the agent away from pgrep -f 'CMD_NAME' as a process-completion poll, and pointing at two safe alternatives.
Background
A live kelos-workers-issue-comment-* pod was found wedged with two orphaned polling shells the agent had written:
bash -c "until [ -z \"\$(pgrep -f 'make verify')\" ]; do sleep 5; done; tail -60 OUT"
bash -c "until [ ! -e /proc/\$(pgrep -f 'go test ./internal/cli/') ]; do sleep 2; done; cat OUT | tail -30"
pgrep -f PAT matches any process whose /proc/*/cmdline contains PAT. The shell running pgrep -f 'make verify' has pgrep -f 'make verify' in its own argv, so pgrep finds the caller itself. The until condition is never empty, the loop runs forever, and the tail after the && never executes — so the BashOutput tool call the agent is awaiting never returns. Claude sits in do_epoll_wait holding pidfds for the wedged shells, the entrypoint waits on claude, and the pod cannot terminate.
The real make verify finished long before (OUT ends with Verification passed). Only the polling shell hangs.
Which issue(s) this PR is related to:
N/A
Special notes for your reviewer:
This is a targeted prompt fix, not a ban on run_in_background: true. Backgrounding is fine; only the self-matching polling form is forbidden. Two safe alternatives are suggested:
- PID capture:
cmd & PID=$!; while kill -0 "$PID" 2>/dev/null; do sleep N; done - Self-exclusion in pgrep:
pgrep -f 'CMD_NAME' | grep -vw $$
The same bullet is added to kelos-workers-agent (only used by the workers spawner) and kelos-dev-agent (shared by kelos-pr-responder, kelos-config-update, kelos-triage, and kelos-squash-commits), so all coding agents pick it up.
Does this PR introduce a user-facing change?
NONE