neuralos 3.3.3 → 3.3.5

package/bin/gybackend.cjs

@@ -389282,6 +389282,233 @@ var RumLedger = class {
 init_infraMonitor();
 init_etwService();
 
+// ../../packages/backend/src/services/dashboard/dashboardPresentation.ts
+var WORK_LIMIT = 8;
+var SITUATION_LIMIT = 8;
+function emptyHint(section, hasRows, message) {
+  if (hasRows) return null;
+  return { section, message };
+}
+function dashboardEmptyHints(state) {
+  const hints = [];
+  const push2 = (section, hasRows, message) => {
+    const h = emptyHint(section, hasRows, message);
+    if (h) hints.push(h);
+  };
+  push2(
+    "hosts",
+    (state.hosts?.length ?? 0) > 0,
+    "No hosts reporting yet. Open a terminal so Monitor can collect CPU/mem/disk, or add a watchdog target."
+  );
+  push2(
+    "slos",
+    (state.slos?.length ?? 0) > 0,
+    "No SLOs defined yet. Define an SLO to track error budget and burn rate."
+  );
+  push2(
+    "uptime",
+    (state.uptime?.length ?? 0) > 0,
+    "No watchdog targets yet. Uptime probes appear here once a host is watched."
+  );
+  push2("incidents", (state.incidents?.length ?? 0) > 0, "No open incidents.");
+  push2(
+    "apm",
+    (state.apm?.bottleneckServices?.length ?? 0) > 0 || (state.apm?.slowestTraces?.length ?? 0) > 0,
+    "No APM spans ingested yet. Model calls and OTLP spans appear here when tracing is on."
+  );
+  push2(
+    "dem",
+    (state.dem?.slowestPages?.length ?? 0) > 0,
+    "No RUM sessions yet. Core Web Vitals appear after DEM beacons are ingested."
+  );
+  push2(
+    "clusters",
+    (state.clusters?.length ?? 0) > 0,
+    "No Kubernetes clusters reporting yet. Run collect_infra / kubectl ingest to populate."
+  );
+  push2(
+    "capacity",
+    (state.capacity?.length ?? 0) > 0,
+    "No capacity forecast yet. Disk-days-to-full needs Monitor disk samples."
+  );
+  return hints;
+}
+function rankSituation(state, extras = {}) {
+  const items = [];
+  for (const u of state.uptime ?? []) {
+    const name = u.target?.name;
+    if (!name) continue;
+    if (u.state === "down") {
+      items.push({
+        id: `down:${name}`,
+        severity: "critical",
+        kind: "host-down",
+        title: `${name} is down`,
+        detail: u.lastError || `${u.consecutiveFailures ?? 0} consecutive probe failure(s)`,
+        action: { type: "open-host", target: name, label: "Open host" }
+      });
+    } else if (u.state === "degraded") {
+      items.push({
+        id: `deg:${name}`,
+        severity: "warning",
+        kind: "host-degraded",
+        title: `${name} is degraded`,
+        detail: u.lastError || "Watchdog reports degraded",
+        action: { type: "open-host", target: name, label: "Open host" }
+      });
+    }
+  }
+  for (const i of state.incidents ?? []) {
+    const sev = String(i.severity || "sev3");
+    const critical = sev === "sev1" || sev === "sev2";
+    items.push({
+      id: `inc:${i.id || i.title || "unknown"}`,
+      severity: critical ? "critical" : "warning",
+      kind: "incident",
+      title: String(i.title || "Open incident"),
+      detail: `${sev} \xB7 ${(i.affected || []).join(", ") || "no affected hosts"}`,
+      action: { type: "open-chat", label: "Ask agent" }
+    });
+  }
+  for (const s of state.slos ?? []) {
+    if (!s.fastBurning) continue;
+    items.push({
+      id: `slo:${s.sloId}`,
+      severity: "critical",
+      kind: "slo-burn",
+      title: `SLO ${s.sloId} is fast-burning`,
+      detail: `burn ${s.burnRate != null ? s.burnRate.toFixed(2) : "\u2014"}x \xB7 budget ${s.errorBudgetRemaining != null ? Math.round(s.errorBudgetRemaining * 100) : "\u2014"}%`,
+      action: { type: "open-connections", target: "playbooks", label: "Playbooks" }
+    });
+  }
+  for (const c of state.capacity ?? []) {
+    if (c.daysToFull != null && c.daysToFull < 7) {
+      items.push({
+        id: `disk:${c.host}`,
+        severity: c.daysToFull < 2 ? "critical" : "warning",
+        kind: "disk-full",
+        title: `${c.host} disk full in ${c.daysToFull.toFixed(1)}d`,
+        detail: `disk ${c.diskPercent != null ? c.diskPercent.toFixed(1) : "\u2014"}%`,
+        action: { type: "open-host", target: c.host, label: "Open host" }
+      });
+    }
+  }
+  for (const cl of state.clusters ?? []) {
+    const crash = cl.crashLoopPods ?? 0;
+    const notReady = cl.notReadyPods ?? 0;
+    if (crash > 0 || notReady > 0) {
+      items.push({
+        id: `k8s:${cl.context || "cluster"}`,
+        severity: crash > 0 ? "critical" : "warning",
+        kind: "cluster-unhealthy",
+        title: `${cl.context || "cluster"} unhealthy`,
+        detail: `${crash} crashloop \xB7 ${notReady} not ready \xB7 ${cl.totalPods ?? 0} pods`
+      });
+    }
+  }
+  for (const p of extras.playbooks ?? []) {
+    if (p.lastRunOk === false) {
+      items.push({
+        id: `pb:${p.id}`,
+        severity: "warning",
+        kind: "playbook-failed",
+        title: `Playbook "${p.name}" last run failed`,
+        detail: p.lastRunAt ? `last run ${p.lastRunAt}` : "last run failed",
+        action: { type: "open-connections", target: "playbooks", label: "Open playbooks" }
+      });
+    }
+  }
+  for (const r of extras.agentRuns ?? []) {
+    if (r.status === "failed" || r.status === "aborted") {
+      items.push({
+        id: `run:${r.runId}`,
+        severity: r.status === "failed" ? "warning" : "info",
+        kind: "agent-failed",
+        title: `Agent run ${r.status}`,
+        detail: r.error || r.inputPreview || r.sessionId,
+        action: { type: "open-chat", target: r.sessionId, label: "Open chat" }
+      });
+    }
+  }
+  const order = { critical: 0, warning: 1, info: 2 };
+  items.sort((a, b) => order[a.severity] - order[b.severity]);
+  const sliced = items.slice(0, SITUATION_LIMIT);
+  if (sliced.length === 0) {
+    const n2 = state.hosts?.length ?? 0;
+    return [
+      {
+        id: "clear",
+        severity: "info",
+        kind: "all-clear",
+        title: n2 > 0 ? `All clear \xB7 ${n2} host${n2 === 1 ? "" : "s"} reporting` : "All clear \xB7 nothing to act on",
+        detail: "No down hosts, fast-burning SLOs, or failed runs."
+      }
+    ];
+  }
+  return sliced;
+}
+function collectWork(extras = {}) {
+  const playbooks = (extras.playbooks ?? []).filter((p) => p.lastRunAt).sort((a, b) => String(b.lastRunAt).localeCompare(String(a.lastRunAt))).slice(0, WORK_LIMIT).map((p) => ({
+    id: p.id,
+    kind: "playbook",
+    name: p.name,
+    ok: p.lastRunOk ?? null,
+    at: p.lastRunAt,
+    detail: p.lastRunOk === false ? "failed" : p.lastRunOk === true ? "ok" : "ran",
+    action: { type: "open-connections", target: "playbooks", label: "Open" }
+  }));
+  const triggers = (extras.triggers ?? []).filter((t) => t.lastFiredAt || (t.fireCount ?? 0) > 0).sort((a, b) => (b.lastFiredAt ?? 0) - (a.lastFiredAt ?? 0)).slice(0, WORK_LIMIT).map((t) => ({
+    id: t.id,
+    kind: "trigger",
+    name: t.name,
+    ok: true,
+    at: t.lastFiredAt,
+    detail: `${t.kind || "trigger"} \xB7 fired ${t.fireCount ?? 0}\xD7`,
+    action: { type: "open-connections", target: "triggers", label: "Open" }
+  }));
+  const scheduledTasks = (extras.scheduledTasks ?? []).filter((t) => t.lastRunAt).sort((a, b) => String(b.lastRunAt).localeCompare(String(a.lastRunAt))).slice(0, WORK_LIMIT).map((t) => ({
+    id: t.id,
+    kind: "scheduled-task",
+    name: t.name,
+    ok: t.enabled === false ? null : true,
+    at: t.lastRunAt,
+    detail: t.enabled === false ? "disabled" : "last run",
+    action: { type: "open-connections", target: "scheduledTasks", label: "Open" }
+  }));
+  const agentRuns = (extras.agentRuns ?? []).slice().sort((a, b) => (b.startedAt || 0) - (a.startedAt || 0)).slice(0, WORK_LIMIT).map((r) => ({
+    id: r.runId,
+    kind: "agent-run",
+    name: r.inputPreview?.slice(0, 60) || r.runId,
+    ok: r.status === "completed" ? true : r.status === "failed" || r.status === "aborted" ? false : null,
+    at: r.endedAt || r.startedAt,
+    detail: r.status + (r.error ? ` \xB7 ${r.error}` : ""),
+    action: { type: "open-chat", target: r.sessionId, label: "Open chat" }
+  }));
+  return { playbooks, triggers, scheduledTasks, agentRuns };
+}
+function situationHeadline(items) {
+  const crit = items.filter((i) => i.severity === "critical" && i.kind !== "all-clear").length;
+  const warn = items.filter((i) => i.severity === "warning").length;
+  if (items.length === 1 && items[0].kind === "all-clear") return items[0].title;
+  const parts = [];
+  if (crit) parts.push(`${crit} critical`);
+  if (warn) parts.push(`${warn} warning`);
+  const top = items[0];
+  return `${parts.join(" \xB7 ") || "attention"} \xB7 ${top.title}`;
+}
+function presentDashboard(state, extras = {}) {
+  const situation = rankSituation(state, extras);
+  return {
+    situation,
+    headline: situationHeadline(situation),
+    work: collectWork(extras),
+    empty: dashboardEmptyHints(state),
+    hostCount: state.hosts?.length ?? 0,
+    downCount: (state.uptime ?? []).filter((u) => u.state === "down").length,
+    degradedCount: (state.uptime ?? []).filter((u) => u.state === "degraded").length
+  };
+}
+
 // ../../packages/backend/src/services/dashboard/dashboardService.ts
 var DEFAULT_LIMIT6 = 10;
 var DashboardService = class {
@@ -389317,7 +389544,7 @@ var DashboardService = class {
     } : { slowestPages: [], poorPages: [] };
     const clusters = d.infraMonitor ? d.infraMonitor.clusters_().slice(0, this.limit) : [];
     const capacity = d.goldenSignals.capacityForecast();
-    return {
+    const fleet = {
       at: this.now(),
       hosts,
       slos,
@@ -389328,6 +389555,20 @@ var DashboardService = class {
       clusters,
       capacity
     };
+    let extras = {};
+    try {
+      extras = this.deps.getExtras?.() ?? {};
+    } catch {
+      extras = {};
+    }
+    const presented = presentDashboard(fleet, extras);
+    return {
+      ...fleet,
+      situation: presented.situation,
+      headline: presented.headline,
+      work: presented.work,
+      empty: presented.empty
+    };
   }
   /** A compact summary line for the gateway/agent (e.g. "3 hosts, 1 degraded, 2 open incidents, 1 SLO fast-burning"). */
   async summary() {
@@ -394742,7 +394983,40 @@ function createObservability(deps) {
     incidentLedger,
     spanLedger,
     rumLedger,
-    infraMonitor
+    infraMonitor,
+    getExtras: () => {
+      const am = deps.automationManager;
+      const playbooks = am.listPlaybooks().map((p) => ({
+        id: p.id,
+        name: p.name,
+        lastRunAt: p.lastRunAt,
+        lastRunOk: p.lastRunOk
+      }));
+      const scheduledTasks = am.listScheduledTasks().map((t) => ({
+        id: t.id,
+        name: t.name,
+        enabled: t.enabled,
+        lastRunAt: t.lastRunAt
+      }));
+      const triggers = am.listTriggers().map((t) => ({
+        id: t.id,
+        name: t.name,
+        enabled: t.enabled,
+        kind: t.kind,
+        lastFiredAt: t.lastFiredAt,
+        fireCount: t.fireCount
+      }));
+      const agentRuns = deps.agentRunLedger.listRuns({ limit: 12 }).map((r) => ({
+        runId: r.runId,
+        sessionId: r.sessionId,
+        status: r.status,
+        error: r.error,
+        startedAt: r.startedAt,
+        endedAt: r.endedAt,
+        inputPreview: r.inputPreview
+      }));
+      return { playbooks, scheduledTasks, triggers, agentRuns };
+    }
   });
   const anomalyDetector = new AnomalyDetector(metricsLedger);
   const earlyWarning = new EarlyWarningService({
@@ -395411,7 +395685,30 @@ function renderDashboardHtml(state, opts = {}) {
       <td>${pct(c.diskPercent)}</td>
       <td>${c.daysToFull !== void 0 ? num2(c.daysToFull, 1) + " days" : "\u2014"}</td>
     </tr>`).join("\n");
-  const section = (id, label, rows, emptyMsg) => live ? `<section><h2>${label}</h2><div id="${id}">${rows ? `<table>${rows}</table>` : `<p class="empty">${esc2(emptyMsg)}</p>`}</div></section>` : rows ? `<section id="${id}"><h2>${label}</h2><table>${rows}</table></section>` : `<section id="${id}"><h2>${label}</h2><p class="empty">${esc2(emptyMsg)}</p></section>`;
+  const emptyMsg = (sectionId, fallback) => state.empty?.find((e) => e.section === sectionId)?.message ?? fallback;
+  const sitItems = state.situation ?? [];
+  const sitHtml = sitItems.map((s) => {
+    const cls = s.severity === "critical" ? "bad" : s.severity === "warning" ? "warn" : "ok";
+    const act = s.action ? `<button class="act" data-action="${esc2(s.action.type)}" data-target="${esc2(s.action.target ?? "")}">${esc2(s.action.label)}</button>` : "";
+    return `<div class="sit ${cls}"><span class="badge ${cls}">${esc2(s.severity)}</span><div><strong>${esc2(s.title)}</strong><div class="dim">${esc2(s.detail)}</div></div>${act}</div>`;
+  }).join("");
+  const workBlock = (title2, items) => {
+    if (!items.length) return "";
+    const rows = items.map((w) => {
+      const badge = w.ok === false ? "bad" : w.ok === true ? "ok" : "mute";
+      const act = w.action ? `<button class="act" data-action="${esc2(w.action.type)}" data-target="${esc2(w.action.target ?? "")}">${esc2(w.action.label)}</button>` : "";
+      return `<tr><td class="host">${esc2(w.name)}</td><td><span class="badge ${badge}">${esc2(w.detail)}</span></td><td>${act}</td></tr>`;
+    }).join("");
+    return `<h3>${esc2(title2)}</h3><table>${rows}</table>`;
+  };
+  const work = state.work;
+  const workHtml = work ? [
+    workBlock("Playbooks", work.playbooks),
+    workBlock("Triggers", work.triggers),
+    workBlock("Scheduled tasks", work.scheduledTasks),
+    workBlock("Agent runs", work.agentRuns)
+  ].join("") : "";
+  const section = (id, label, rows, emptyMsgText) => live ? `<section><h2>${label}</h2><div id="${id}">${rows ? `<table>${rows}</table>` : `<p class="empty">${esc2(emptyMsgText)}</p>`}</div></section>` : rows ? `<section id="${id}"><h2>${label}</h2><table>${rows}</table></section>` : `<section id="${id}"><h2>${label}</h2><p class="empty">${esc2(emptyMsgText)}</p></section>`;
   return `<!DOCTYPE html>
 <html lang="en">
 <head>
@@ -395451,6 +395748,16 @@ ${refresh > 0 && !live ? `<meta http-equiv="refresh" content="${refresh}">` : ""
   .row-bad td { background: rgba(255,93,126,0.05); }
   .empty { color: var(--muted); font-size: 12px; padding: 8px 0; }
   .span2 { grid-column: span 2; }
+  .sit-strip { display: flex; flex-direction: column; gap: 8px; }
+  .sit { display: flex; align-items: flex-start; gap: 10px; padding: 8px 10px; border-radius: 8px; border: 1px solid var(--border); }
+  .sit.bad { border-color: rgba(255,93,126,0.35); background: rgba(255,93,126,0.06); }
+  .sit.warn { border-color: rgba(255,196,77,0.3); background: rgba(255,196,77,0.05); }
+  .sit.ok { border-color: rgba(61,220,151,0.25); }
+  .sit strong { display: block; font-size: 13px; }
+  .act { margin-left: auto; background: transparent; color: var(--accent); border: 1px solid rgba(79,216,232,0.35); border-radius: 6px; padding: 3px 8px; font-size: 11px; cursor: pointer; white-space: nowrap; }
+  .act:hover { background: rgba(79,216,232,0.12); }
+  section h3 { font-size: 11px; color: var(--muted); margin: 10px 0 6px; text-transform: uppercase; letter-spacing: 0.06em; }
+  .filter { margin: 0 0 10px; padding: 6px 10px; width: 100%; max-width: 320px; background: var(--panel2); color: var(--fg); border: 1px solid var(--border); border-radius: 8px; }
   footer { color: var(--muted); font-size: 11px; padding: 14px 26px 30px; }
   @media (max-width: 900px) { .grid { grid-template-columns: 1fr; } .span2 { grid-column: span 1; } }
 </style>
@@ -395462,25 +395769,36 @@ ${refresh > 0 && !live ? `<meta http-equiv="refresh" content="${refresh}">` : ""
   <span class="live" id="live-indicator">LIVE \xB7 updated ${esc2(new Date(state.at).toISOString().slice(11, 19))} UTC</span>
 </header>
 <main>
+  <section class="span2" id="situation-wrap">
+    <h2>Situation</h2>
+    <div class="sub dim" id="headline">${esc2(state.headline || "")}</div>
+    <div class="sit-strip" id="situation">${sitHtml || '<p class="empty">All clear.</p>'}</div>
+  </section>
   <div class="grid">
     <section class="span2">
       <h2>Fleet health</h2>
+      <input class="filter" id="host-filter" type="search" placeholder="Filter hosts\u2026" />
       <div id="fleet">
       <table>
         <tr><th>host</th><th>state</th><th>cpu</th><th>mem</th><th>disk</th><th>cpu trend</th><th>disk full in</th></tr>
-        ${hostRows || '<tr><td class="empty">No hosts reporting yet.</td></tr>'}
+        ${hostRows || `<tr><td class="empty">${esc2(emptyMsg("hosts", "No hosts reporting yet. Open a terminal so Monitor can collect CPU/mem/disk, or add a watchdog target."))}</td></tr>`}
       </table>
       </div>
     </section>
 
-    ${section("slo", "SLO board", sloRows ? `<tr><th>slo</th><th>sli</th><th>burn rate</th><th>budget</th><th>status</th></tr>${sloRows}` : "", "No SLOs defined yet.")}
-    ${section("uptime", "Uptime", upRows ? `<tr><th>host</th><th>state</th><th>failures</th><th>latency</th><th>error</th></tr>${upRows}` : "", "No watchdog targets yet.")}
-    ${section("incidents", "Open incidents", incRows ? `<tr><th>incident</th><th>sev</th><th>status</th><th>affected</th><th>rca</th></tr>${incRows}` : "", "No open incidents.")}
-    ${section("apm-svc", "APM \xB7 bottleneck services", apmSvcRows ? `<tr><th>service</th><th>spans</th><th>errors</th><th>error rate</th><th>p95</th></tr>${apmSvcRows}` : "", "No APM spans ingested yet.")}
-    ${section("apm-trace", "APM \xB7 slowest traces", apmTraceRows ? `<tr><th>trace</th><th>root</th><th>spans</th><th>duration</th><th>status</th></tr>${apmTraceRows}` : "", "No traces yet.")}
-    ${section("dem", "DEM \xB7 slowest pages (Core Web Vitals)", demRows ? `<tr><th>page</th><th>sessions</th><th>p75 lcp</th><th>p75 inp</th><th>error rate</th></tr>${demRows}` : "", "No RUM sessions yet.")}
-    ${section("clusters", "Kubernetes / cloud clusters", clusterRows ? `<tr><th>context</th><th>pods</th><th>not ready</th><th>crashloop</th><th>restarts</th><th>nodes</th></tr>${clusterRows}` : "", "No clusters reporting yet.")}
-    ${section("capacity", "Capacity forecast", capRows ? `<tr><th>host</th><th>disk</th><th>full in</th></tr>${capRows}` : "", "No capacity data yet.")}
+    <section class="span2" id="work-wrap">
+      <h2>RTerm work</h2>
+      <div id="work">${workHtml || '<p class="empty">No playbook runs, trigger fires, scheduled tasks, or agent runs yet.</p>'}</div>
+    </section>
+
+    ${section("slo", "SLO board", sloRows ? `<tr><th>slo</th><th>sli</th><th>burn rate</th><th>budget</th><th>status</th></tr>${sloRows}` : "", emptyMsg("slos", "No SLOs defined yet. Define an SLO to track error budget and burn rate."))}
+    ${section("uptime", "Uptime", upRows ? `<tr><th>host</th><th>state</th><th>failures</th><th>latency</th><th>error</th></tr>${upRows}` : "", emptyMsg("uptime", "No watchdog targets yet. Uptime probes appear here once a host is watched."))}
+    ${section("incidents", "Open incidents", incRows ? `<tr><th>incident</th><th>sev</th><th>status</th><th>affected</th><th>rca</th></tr>${incRows}` : "", emptyMsg("incidents", "No open incidents."))}
+    ${section("apm-svc", "APM \xB7 bottleneck services", apmSvcRows ? `<tr><th>service</th><th>spans</th><th>errors</th><th>error rate</th><th>p95</th></tr>${apmSvcRows}` : "", emptyMsg("apm", "No APM spans ingested yet. Model calls and OTLP spans appear here when tracing is on."))}
+    ${section("apm-trace", "APM \xB7 slowest traces", apmTraceRows ? `<tr><th>trace</th><th>root</th><th>spans</th><th>duration</th><th>status</th></tr>${apmTraceRows}` : "", emptyMsg("apm", "No traces yet."))}
+    ${section("dem", "DEM \xB7 slowest pages (Core Web Vitals)", demRows ? `<tr><th>page</th><th>sessions</th><th>p75 lcp</th><th>p75 inp</th><th>error rate</th></tr>${demRows}` : "", emptyMsg("dem", "No RUM sessions yet. Core Web Vitals appear after DEM beacons are ingested."))}
+    ${section("clusters", "Kubernetes / cloud clusters", clusterRows ? `<tr><th>context</th><th>pods</th><th>not ready</th><th>crashloop</th><th>restarts</th><th>nodes</th></tr>${clusterRows}` : "", emptyMsg("clusters", "No Kubernetes clusters reporting yet. Run collect_infra / kubectl ingest to populate."))}
+    ${section("capacity", "Capacity forecast", capRows ? `<tr><th>host</th><th>disk</th><th>full in</th></tr>${capRows}` : "", emptyMsg("capacity", "No capacity forecast yet. Disk-days-to-full needs Monitor disk samples."))}
   </div>
 </main>
 <footer>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "neuralos",
-  "version": "3.3.3",
+  "version": "3.3.5",
   "description": "AI-native terminal & agentic-AI operations platform for Forward Deployed Engineers & SREs: AIOps closed-loop remediation, AI SRE, self-healing infrastructure, runbook automation, ChatOps; executes over SSH/WinRM/serial under policy with tamper-evident audit.",
   "keywords": [
     "forward-deployed-engineer",