Skip to content

Business operating system for Claude Code — 57 skills, 21 agents, smart daemon. Unified inbox (WhatsApp/Email/Slack/Telegram), autonomous PR merge, full-AWS monitoring, revenue (Stripe+RevenueCat), e-commerce (Shopify), marketing (Klaviyo/Meta/GA4), voice (Bland/ElevenLabs), APM (Datadog/NewRelic/OTEL), YOLO mode.

feat(ops-daemon): kill stuck `gh ... --watch` processes (REST rate-limit hygiene) (#214)

Aurora Capital70eb78d34236
diff --git a/claude-ops/scripts/ops-daemon.sh b/claude-ops/scripts/ops-daemon.sh
index fd91992d..c4e88c24 100755
--- a/claude-ops/scripts/ops-daemon.sh
+++ b/claude-ops/scripts/ops-daemon.sh
@@ -1087,6 +1087,62 @@ PY
date +%s > "$LAST_CHECK"
}
+# ── Kill stuck gh polling processes (REST rate-limit hygiene) ────────────
+# `gh pr checks <N> --watch` and `gh run watch <N>` poll the GitHub REST API
+# every ~2s indefinitely. When left behind by an interrupted skill or a
+# forgotten foreground command, a single watcher can burn 1800 calls/hr —
+# half the 5000/hr personal token bucket. This kills any gh poller older
+# than MAX_AGE_SECONDS (default 30 min) so the REST quota stays healthy
+# for interactive work.
+#
+# Scope: only matches `gh.*--watch` (the polling pattern). One-shot kills
+# (`gh pr merge`, `gh api`) are not affected. Self-throttled to 5 min.
+kill_stuck_gh_watchers() {
+ local LAST_CHECK="$DATA_DIR/cache/.gh_watcher_kill_ts"
+ local now; now=$(date +%s)
+ mkdir -p "$DATA_DIR/cache" 2>/dev/null || true
+ if [[ -f "$LAST_CHECK" ]]; then
+ local last; last=$(cat "$LAST_CHECK" 2>/dev/null || echo 0)
+ if (( now - last < 300 )); then return 0; fi
+ fi
+
+ local MAX_AGE_SECONDS="${OPS_GH_WATCHER_MAX_AGE:-1800}"
+ local killed=0
+
+ # ps -eo etime format: "MM:SS" | "HH:MM:SS" | "DD-HH:MM:SS"
+ # Parse via awk to seconds, kill anything older than threshold.
+ while IFS= read -r line; do
+ [[ -z "$line" ]] && continue
+ local pid="${line%% *}"
+ [[ -z "$pid" ]] && continue
+ log "WATCHER-KILL: terminating stuck gh poller (PID $pid, age >${MAX_AGE_SECONDS}s)"
+ kill -TERM "$pid" 2>/dev/null || true
+ killed=$((killed + 1))
+ done < <(ps -eo pid,etime,command 2>/dev/null | awk -v max="$MAX_AGE_SECONDS" '
+ # Match gh ... --watch (excluding our own awk and grep noise)
+ /gh .*--watch/ && !/awk/ && !/grep/ {
+ et = $2
+ secs = 0
+ # Strip days prefix "DD-"
+ if (et ~ /-/) {
+ n = split(et, a, "-")
+ secs += a[1] * 86400
+ et = a[2]
+ }
+ # Parse remaining HH:MM:SS or MM:SS
+ n = split(et, b, ":")
+ if (n == 3) secs += b[1]*3600 + b[2]*60 + b[3]
+ else if (n == 2) secs += b[1]*60 + b[2]
+ if (secs > max) print $1
+ }
+ ')
+
+ if (( killed > 0 )); then
+ log "WATCHER-KILL: killed $killed stuck gh poller process(es)"
+ fi
+ date +%s > "$LAST_CHECK"
+}
+
# ── Phase 16: credential expiry (offline) ────────────────────────────────
# Reads *_expires_at / *_created_at from preferences.json and warns:
# - expires within WARN_DAYS (7)
@@ -1730,6 +1786,7 @@ while true; do
# ── Phase 16: infrastructure hardening ───────────────────────────────────
check_rate_limits # Every 5 min: reset rollover windows + warnings
+ kill_stuck_gh_watchers # Every 5 min: TERM gh ... --watch processes >30 min old
check_credential_expiry # Hourly: offline check of *_expires_at / *_created_at
write_daemon_health