@dudousxd/nestjs-agent-telescope 0.8.60 → 0.9.0

package/dist/index.cjs

@@ -33,7 +33,9 @@ __export(index_exports, {
   agentRecentThreadsTableProvider: () => agentRecentThreadsTableProvider,
   agentRecentToolCallsTableProvider: () => agentRecentToolCallsTableProvider,
   agentRunErrorsProvider: () => agentRunErrorsProvider,
+  agentRunFailureCodesProvider: () => agentRunFailureCodesProvider,
   agentRunsByAgentTableProvider: () => agentRunsByAgentTableProvider,
+  agentRunsCancelledProvider: () => agentRunsCancelledProvider,
   agentRunsDurationProvider: () => agentRunsDurationProvider,
   agentRunsFailedProvider: () => agentRunsFailedProvider,
   agentRunsProvider: () => agentRunsProvider,
@@ -41,6 +43,8 @@ __export(index_exports, {
   agentRunsSuccessRateProvider: () => agentRunsSuccessRateProvider,
   agentRunsTotalProvider: () => agentRunsTotalProvider,
   agentRunsTrendProvider: () => agentRunsTrendProvider,
+  agentRunsTtftDistributionProvider: () => agentRunsTtftDistributionProvider,
+  agentRunsTtftProvider: () => agentRunsTtftProvider,
   agentSpendByActorProvider: () => agentSpendByActorProvider,
   agentSpendByModelProvider: () => agentSpendByModelProvider,
   agentSpendTotalProvider: () => agentSpendTotalProvider,
@@ -72,6 +76,7 @@ __export(index_exports, {
   toActorSpendRows: () => toActorSpendRows,
   toActorSpendSegments: () => toActorSpendSegments,
   toCollectionRows: () => toCollectionRows,
+  toFailureCodeSegments: () => toFailureCodeSegments,
   toLatencyDistribution: () => toLatencyDistribution,
   toModelSpendRows: () => toModelSpendRows,
   toModelSpendSegments: () => toModelSpendSegments,
@@ -90,6 +95,7 @@ __export(index_exports, {
   toStoreSegments: () => toStoreSegments,
   toThreadSpendRows: () => toThreadSpendRows,
   toToolStatTableRows: () => toToolStatTableRows,
+  toTtftDistribution: () => toTtftDistribution,
   toUsageTrendRows: () => toUsageTrendRows,
   totalCostUsd: () => totalCostUsd,
   totalTokens: () => totalTokens,
@@ -340,6 +346,43 @@ function agentDashboard(opts = {}) {
           },
           // Two duration stats above, not a `distribution` panel: RunMetrics carries only p50/p95
           // (no raw samples to bucket), so a histogram here would be a permanently-empty box.
+          {
+            kind: "stat",
+            title: "TTFT p50",
+            data: {
+              provider: "agent.runs.ttft",
+              query: {
+                metric: "p50"
+              }
+            },
+            format: "duration"
+          },
+          {
+            kind: "stat",
+            title: "TTFT p95",
+            data: {
+              provider: "agent.runs.ttft",
+              query: {
+                metric: "p95"
+              }
+            },
+            format: "duration"
+          },
+          {
+            // The one histogram in this section: `run.finished` diagnostics carry each run's own
+            // `ttftMs`, so the watcher-recorded entries DO have samples to bucket (over what
+            // Telescope's storage still holds, not the governance range).
+            kind: "distribution",
+            title: "Time to first token",
+            data: {
+              provider: "agent.runs.ttftDistribution"
+            },
+            markers: [
+              "p50",
+              "p95"
+            ],
+            format: "duration"
+          },
           {
             kind: "timeseries",
             title: "Runs & failures",
@@ -387,6 +430,30 @@ function agentDashboard(opts = {}) {
           }
         ]
       },
+      {
+        // Live, from the watcher-recorded diagnostics rather than the run rows: what failed and
+        // what was stopped, readable on a host that binds no governance read-model. A cancel is
+        // its own number — a user pressing Stop is the control working, not a failure.
+        title: "Run outcomes",
+        cols: 2,
+        panels: [
+          {
+            kind: "breakdown",
+            title: "Failures by code",
+            data: {
+              provider: "agent.runs.failureCodes"
+            },
+            style: "donut"
+          },
+          {
+            kind: "stat",
+            title: "Cancelled",
+            data: {
+              provider: "agent.runs.cancelled"
+            }
+          }
+        ]
+      },
       {
         // The paged list tables: tool calls, threads, runs — each reads the paged SPI
         // (toolCallsPage/threadsPage/runsPage via `agent.*.paged`) and opts into the renderer's
@@ -783,6 +850,82 @@ function ofEvent(entries, event) {
   return entries.filter((entry) => entry.content?.event === event);
 }
 __name(ofEvent, "ofEvent");
+function firstToolCalls(entries) {
+  return ofEvent(entries, "tool-call").filter((entry) => entry.content?.replayed !== true);
+}
+__name(firstToolCalls, "firstToolCalls");
+function percentileOf(samples, fraction) {
+  if (samples.length === 0) {
+    return null;
+  }
+  const sorted = [
+    ...samples
+  ].sort((a, b) => a - b);
+  const rank = Math.ceil(fraction * sorted.length);
+  return sorted[Math.min(Math.max(rank, 1), sorted.length) - 1] ?? null;
+}
+__name(percentileOf, "percentileOf");
+var TTFT_EDGES = [
+  250,
+  500,
+  1e3,
+  2e3,
+  4e3,
+  8e3,
+  16e3
+];
+function ttftLabel(index) {
+  const upper = TTFT_EDGES[index];
+  if (upper === void 0) {
+    return `\u2265${TTFT_EDGES[TTFT_EDGES.length - 1] ?? 0}ms`;
+  }
+  return `${index === 0 ? 0 : TTFT_EDGES[index - 1] ?? 0}\u2013${upper}ms`;
+}
+__name(ttftLabel, "ttftLabel");
+function toTtftDistribution(entries) {
+  const samples = [];
+  const counts = new Array(TTFT_EDGES.length + 1).fill(0);
+  for (const entry of ofEvent(entries, "run.finished")) {
+    const ttftMs = entry.content?.ttftMs;
+    if (typeof ttftMs !== "number" || entry.content?.parentRunId !== void 0) {
+      continue;
+    }
+    samples.push(ttftMs);
+    const found = TTFT_EDGES.findIndex((edge) => ttftMs < edge);
+    const index = found === -1 ? TTFT_EDGES.length : found;
+    counts[index] = (counts[index] ?? 0) + 1;
+  }
+  const buckets = counts.map((count, index) => ({
+    label: ttftLabel(index),
+    count
+  }));
+  const p50 = percentileOf(samples, 0.5);
+  const p95 = percentileOf(samples, 0.95);
+  return {
+    buckets,
+    ...p50 !== null ? {
+      p50
+    } : {},
+    ...p95 !== null ? {
+      p95
+    } : {}
+  };
+}
+__name(toTtftDistribution, "toTtftDistribution");
+function toFailureCodeSegments(entries) {
+  const counts = /* @__PURE__ */ new Map();
+  for (const entry of ofEvent(entries, "run.failed")) {
+    const code = entry.content?.code ?? "unknown";
+    counts.set(code, (counts.get(code) ?? 0) + 1);
+  }
+  return [
+    ...counts.entries()
+  ].map(([label, value]) => ({
+    label,
+    value
+  })).sort((a, b) => b.value - a.value);
+}
+__name(toFailureCodeSegments, "toFailureCodeSegments");
 function agentRunsProvider() {
   return {
     name: "agent.runs",
@@ -812,7 +955,7 @@ function agentToolsProvider() {
   return {
     name: "agent.tools",
     async resolve(_query, ctx) {
-      const calls = ofEvent(await fetchEntries(ctx), "tool-call").slice(-50).reverse().map((entry) => ({
+      const calls = firstToolCalls(await fetchEntries(ctx)).slice(-50).reverse().map((entry) => ({
         toolName: entry.content?.toolName ?? "",
         toolType: entry.content?.toolType ?? "",
         status: entry.content?.status ?? "",
@@ -829,7 +972,7 @@ function agentToolStatusProvider() {
   return {
     name: "agent.toolStatus",
     async resolve(_query, ctx) {
-      const calls = ofEvent(await fetchEntries(ctx), "tool-call");
+      const calls = firstToolCalls(await fetchEntries(ctx));
       const counts = /* @__PURE__ */ new Map();
       for (const call of calls) {
         const status = call.content?.status ?? "unknown";
@@ -847,6 +990,37 @@ function agentToolStatusProvider() {
   };
 }
 __name(agentToolStatusProvider, "agentToolStatusProvider");
+function agentRunsTtftDistributionProvider() {
+  return {
+    name: "agent.runs.ttftDistribution",
+    async resolve(_query, ctx) {
+      return toTtftDistribution(await fetchEntries(ctx));
+    }
+  };
+}
+__name(agentRunsTtftDistributionProvider, "agentRunsTtftDistributionProvider");
+function agentRunFailureCodesProvider() {
+  return {
+    name: "agent.runs.failureCodes",
+    async resolve(_query, ctx) {
+      return {
+        segments: toFailureCodeSegments(await fetchEntries(ctx))
+      };
+    }
+  };
+}
+__name(agentRunFailureCodesProvider, "agentRunFailureCodesProvider");
+function agentRunsCancelledProvider() {
+  return {
+    name: "agent.runs.cancelled",
+    async resolve(_query, ctx) {
+      return {
+        value: ofEvent(await fetchEntries(ctx), "run.cancelled").length
+      };
+    }
+  };
+}
+__name(agentRunsCancelledProvider, "agentRunsCancelledProvider");
 
 // src/agent-governance-providers.ts
 var import_nestjs_agent_core = require("@dudousxd/nestjs-agent-core");
@@ -1004,6 +1178,9 @@ function toRecentRunTableRows(rows) {
     agentName: row.agentName ?? NO_VALUE,
     status: row.status,
     durationMs: row.durationMs ?? NO_VALUE,
+    ttftMs: row.ttftMs ?? NO_VALUE,
+    costUsd: row.costUsd ?? NO_VALUE,
+    modelId: row.modelId ?? NO_VALUE,
     retries: row.retries,
     errorCode: row.errorCode ?? NO_VALUE,
     errorMessage: capErrorMessage(row.errorMessage) ?? NO_VALUE,
@@ -1051,7 +1228,9 @@ var EMPTY_RUN_METRICS = {
   successRate: 0,
   retries: 0,
   durationP50Ms: null,
-  durationP95Ms: null
+  durationP95Ms: null,
+  ttftP50Ms: null,
+  ttftP95Ms: null
 };
 function governanceStatProvider(name, fetch, format) {
   return {
@@ -1161,6 +1340,20 @@ function agentRunsDurationProvider() {
   };
 }
 __name(agentRunsDurationProvider, "agentRunsDurationProvider");
+function agentRunsTtftProvider() {
+  return {
+    name: "agent.runs.ttft",
+    async resolve(query, ctx) {
+      const queries = resolveGovernanceQueries(ctx);
+      const metrics = queries ? await queries.runMetrics(resolveRange(query)) : EMPTY_RUN_METRICS;
+      const value = query?.metric === "p95" ? metrics.ttftP95Ms : metrics.ttftP50Ms;
+      return {
+        value: value ?? 0
+      };
+    }
+  };
+}
+__name(agentRunsTtftProvider, "agentRunsTtftProvider");
 function agentRunsByAgentTableProvider() {
   return governanceStatProvider("agent.runs.byAgent", (queries, range) => queries.runsByAgent(range), (rows) => ({
     rows: toRunAgentTableRows(rows)
@@ -1829,6 +2022,10 @@ function agentTelescopeExtension(opts = {}) {
       agentRunsFailedProvider(),
       agentRunsRetriesProvider(),
       agentRunsDurationProvider(),
+      agentRunsTtftProvider(),
+      agentRunsTtftDistributionProvider(),
+      agentRunFailureCodesProvider(),
+      agentRunsCancelledProvider(),
       agentRunsByAgentTableProvider(),
       agentRunErrorsProvider(),
       agentRunsTrendProvider(),
@@ -1870,7 +2067,9 @@ __name(agentTelescopeExtension, "agentTelescopeExtension");
   agentRecentThreadsTableProvider,
   agentRecentToolCallsTableProvider,
   agentRunErrorsProvider,
+  agentRunFailureCodesProvider,
   agentRunsByAgentTableProvider,
+  agentRunsCancelledProvider,
   agentRunsDurationProvider,
   agentRunsFailedProvider,
   agentRunsProvider,
@@ -1878,6 +2077,8 @@ __name(agentTelescopeExtension, "agentTelescopeExtension");
   agentRunsSuccessRateProvider,
   agentRunsTotalProvider,
   agentRunsTrendProvider,
+  agentRunsTtftDistributionProvider,
+  agentRunsTtftProvider,
   agentSpendByActorProvider,
   agentSpendByModelProvider,
   agentSpendTotalProvider,
@@ -1909,6 +2110,7 @@ __name(agentTelescopeExtension, "agentTelescopeExtension");
   toActorSpendRows,
   toActorSpendSegments,
   toCollectionRows,
+  toFailureCodeSegments,
   toLatencyDistribution,
   toModelSpendRows,
   toModelSpendSegments,
@@ -1927,6 +2129,7 @@ __name(agentTelescopeExtension, "agentTelescopeExtension");
   toStoreSegments,
   toThreadSpendRows,
   toToolStatTableRows,
+  toTtftDistribution,
   toUsageTrendRows,
   totalCostUsd,
   totalTokens,