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.
Shell agent-teamsai-agentai-automationanthropicautonomous-agentawsbusiness-operationsclaudeclaude-codeclaude-code-plugindevopsecommerceinbox-zeroklaviyomarketingmcpmorning-briefingpr-automationshopifyterminal
fix(ops-daemon): honor health_check probe — stop whatsapp-bridge restart churn (#485)
lifecycle-innovations-ops[bot]c86734eaeded
diff --git a/claude-ops/scripts/ops-daemon.sh b/claude-ops/scripts/ops-daemon.shindex 720d1177..1532ffd5 100755--- a/claude-ops/scripts/ops-daemon.sh+++ b/claude-ops/scripts/ops-daemon.sh@@ -111,6 +111,7 @@ declare -A SERVICE_LAST_SUCCESS # service name → ISO-8601 UTC of last cleandeclare -A SERVICE_ERROR_COUNT # service name → cumulative errors since daemon startdeclare -A SERVICE_LAST_ERROR # service name → last error message (truncated)declare -A SERVICE_NOTIFIED_CRASH # service name → set to 1 after persistent-failure notify (deduped)+declare -A SERVICE_HEALTHCHECK_CMD # service name → cached config `health_check` liveness probeRESTART_COOLDOWN=1800 # Reset restart counter after 30 min of stabilityACTION_NEEDED="null"@@ -147,6 +148,43 @@ for name, cfg in data.get('services', {}).items():" 2>/dev/null || true}+# Return the configured command-based liveness probe for a service, if any+# (cached for the daemon's lifetime — the config is static at runtime).+# When a service defines `health_check`, that probe — not the tracked PID —+# is authoritative for liveness. This matters for services whose `command` is+# an ephemeral launcher wrapper that exits the moment the real, externally+# supervised process is confirmed up: e.g. whatsapp-bridge-up.sh returns as+# soon as :8080 is LISTENing under whatsapp-bridge.service. Tracking that+# already-exited wrapper PID otherwise caused a perpetual false dead→restart+# churn (HEALTH: ... is dead → RESTART attempt N/10) for a perfectly healthy+# systemd-owned bridge, eventually tripping max_restarts + a false crash alert.+_service_health_cmd() {+ local name="$1"+ if [[ -z "${SERVICE_HEALTHCHECK_CMD[$name]+set}" ]]; then+ # Probe the config directly (not via get_service_field, whose `|| true`+ # masks parse failure). Cache ONLY on a clean parse: a transient python3+ # spawn / JSON-decode failure (e.g. config mid-rewrite) must not get cached+ # as "no health_check" — that would permanently disable the probe and+ # silently revert the service to the churny PID-liveness path. A genuine+ # empty result (service has no health_check) parses cleanly and is cached.+ local val+ if val=$(python3 -c "+import json, sys+data = json.load(open('$SERVICES_CONFIG'))+v = data.get('services', {}).get('$name', {}).get('health_check', '')+sys.stdout.write(v if v is not None else '')+" 2>/dev/null); then+ SERVICE_HEALTHCHECK_CMD[$name]="$val"+ else+ # Transient failure — return empty for this round WITHOUT caching; retry+ # on the next loop so the probe self-heals once the config is readable.+ printf ''+ return 0+ fi+ fi+ printf '%s' "${SERVICE_HEALTHCHECK_CMD[$name]}"+}+# ── Cron helpers ─────────────────────────────────────────────────────────# Parse a simple cron expression (*/N * * * *) and return next epoch.# Only supports */N minute fields — sufficient for daemon use cases.@@ -255,13 +293,37 @@ check_health() {health_path=$(get_service_field "$name" "health_file")health_path="${health_path/#\~/$HOME}"- # Check if PID is alive- local pid="${SERVICE_PIDS[$name]:-}"- if [[ -n "$pid" ]] && ! kill -0 "$pid" 2>/dev/null; then- log "HEALTH: $name pid=$pid is dead"- SERVICE_STATUS["$name"]="dead"- SERVICE_PIDS["$name"]=""- return 1+ # Liveness check. A configured `health_check` command (e.g. probing :8080)+ # is authoritative and overrides PID tracking: services launched via an+ # ephemeral wrapper (whatsapp-bridge-up.sh) leave a PID that exits the moment+ # the real systemd/launchd-owned process is up, so tracking it would falsely+ # report "dead" on every loop and churn restarts.+ local health_cmd+ health_cmd=$(_service_health_cmd "$name")+ if [[ -n "$health_cmd" ]]; then+ if eval "$health_cmd" >/dev/null 2>&1; then+ # Real process is up; the ephemeral launcher PID is no longer meaningful.+ SERVICE_PIDS["$name"]=""+ SERVICE_STATUS["$name"]="running"+ SERVICE_LAST_HEALTH["$name"]="ok"+ if [[ -n "${SERVICE_NOTIFIED_CRASH[$name]:-}" ]]; then+ unset "SERVICE_NOTIFIED_CRASH[$name]"+ fi+ else+ log "HEALTH: $name health_check failed ($health_cmd) — treating as dead"+ SERVICE_STATUS["$name"]="dead"+ SERVICE_PIDS["$name"]=""+ return 1+ fi+ else+ # Legacy PID-liveness check (services with no health_check probe).+ local pid="${SERVICE_PIDS[$name]:-}"+ if [[ -n "$pid" ]] && ! kill -0 "$pid" 2>/dev/null; then+ log "HEALTH: $name pid=$pid is dead"+ SERVICE_STATUS["$name"]="dead"+ SERVICE_PIDS["$name"]=""+ return 1+ fifi# Read health file if present@@ -388,7 +450,15 @@ print(json.dumps(sys.argv[1]))services_json+="\"$name\": {\"status\": \"$status\", \"next_run\": \"$next_iso\", \"last_run\": \"$last_iso\", \"latency_ms\": $latency_ms, \"last_success\": \"$last_success\", \"error_count\": $error_count, \"last_error\": $last_error_json}"elselocal pid_val- if [[ "$pid" == "null" ]] || [[ -z "$pid" ]]; then+ local hc_cmd+ hc_cmd=$(_service_health_cmd "$name")+ if [[ -n "$hc_cmd" ]]; then+ # Command-supervised service: liveness was already determined by+ # check_health via the configured probe. The tracked PID is an+ # ephemeral launcher and not meaningful — report status as-is with a+ # null pid; do NOT downgrade running→dead on a dead wrapper PID.+ pid_val="null"+ elif [[ "$pid" == "null" ]] || [[ -z "$pid" ]]; thenpid_val="null"# No PID tracked — status cannot be "running"if [[ "$status" == "running" ]]; then