Metrics

AgentStack Native Metrics Endpoint (single source of truth for admin monitoring).

Response contract (all keys present; values may be null/empty/error on failure):

• timestamp (str), service (str), version (str), metrics_source (str), execution_sample_limit (int)
• system: { cpu_percent?, memory?: { total_gb, used_gb, available_gb, percent }, disk?: {...}, error? }
• execution_logs: { [service_name]: stats | { error } }; stats may include outcome_unknown, p50_duration_ms, p95_duration_ms, top_operations, by_log_source, scope_note
• execution_logs_meta: { methodology, diagnostic_sql }
• cache: { [cache_name]: { service, role?, entries, memory_mb?, ... } }
• database: { total_connections?, active_connections?, idle_connections? } | { error }
• uptime_seconds (float), network: { rx, tx }
• services: [{ name, status, uptime, responseTime, lastChecked }]
• alerts: [{ id, level, title, message, timestamp, resolved }]
• performance: { active_session_slots, recent_session_logs_sample_size, recent_session_logs_error_rate_percent, avg_response_time_ms }
• managed_organism: { active_cells, draining_count, by_activity, schedule_heap_pending, schedule_driver_running, runtime_started, io_db_straggler_count, io_db_max_workers, io_db_shutdown } | { error } — aggregates only (no cell names; safe for public metrics)

On collection errors, individual blocks include "error" key; endpoint does not raise.