agent-inspect 6.7.2 → 6.7.4

package/packages/core/dist/advanced.mjs

@@ -1,7 +1,7 @@
-import { runTraceChecks, createRunStatusRule, createToolUsageRule, createRunDurationRule, createLlmUsageRule, createObservedOutcomeRule } from './chunk-EWHN6NA4.mjs';
+import { runTraceChecks, createRunStatusRule, createToolUsageRule, createRunDurationRule, createLlmUsageRule, createObservedOutcomeRule } from './chunk-EQVXDGGU.mjs';
 export { MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, TERMINAL_INDENT, TraceDirectory, createInspector, createInspectorRuntime, formatTerminalName, getCurrentContext, getCurrentCorrelationMetadata, getCurrentDepth, getCurrentRunId, getCurrentRunName, getCurrentStepId, getIndent, getParentStepId, getTraceDirFromContext, getTraceSafetyFromContext, hasActiveContext, isAgentInspectEnabled, isSilentContext, maybeInspectRun, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, renderErrorLine, renderRunSummary, renderStepLine, resolveTraceDir, runWithContext, runWithStepContext } from './chunk-ZWA5RWMV.mjs';
-import { buildRunSummary, extractMetadata } from './chunk-3USJVBLA.mjs';
-export { buildRunSummary, buildRunTimeline, buildRunWhatSummary, extractMetadata, renderRunWhat, renderTimeline } from './chunk-3USJVBLA.mjs';
+import { filterTraces, extractMetadata, buildRunSummary } from './chunk-KDDU2R5P.mjs';
+export { buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildTraceStats, extractMetadata, filterTraces, renderRunWhat, renderTimeline, renderTraceStats } from './chunk-KDDU2R5P.mjs';
 import { parseObservationFilter, extractOutcomesFromTraceEvents, extractOutcomesFromPersistedEvents } from './chunk-VFO76UH3.mjs';
 export { extractOutcomesFromPersistedEvents, extractOutcomesFromTraceEvents } from './chunk-VFO76UH3.mjs';
 import './chunk-BT7CATSD.mjs';
@@ -9,11 +9,11 @@ import { escapeHtml } from './chunk-CAWPF22J.mjs';
 import { resolveRedactionProfile, readTraceEventsFromFile } from './chunk-ULCGIRTC.mjs';
 export { DEFAULT_MAX_EVENT_BYTES, DEFAULT_MAX_METADATA_VALUE_LENGTH, DEFAULT_MAX_PREVIEW_LENGTH, getRunIdFromTraceFileName, initializeTraceFile, listTraceFiles, prepareMetadataForDisk, prepareTraceEventForDisk, readTraceEvents, readTraceFile, resolveRedactionProfile, resolveTraceSafetyOptions, serializeEvent, validateEvent, writeTraceEvent } from './chunk-ULCGIRTC.mjs';
 import { Redactor } from './chunk-VU6O5QAH.mjs';
-import { openTrace } from './chunk-X2FLDF7M.mjs';
-import './chunk-KNYL56KZ.mjs';
+import { openTrace } from './chunk-CTA6XNDP.mjs';
+import './chunk-UOPVGCOB.mjs';
 import './chunk-S4YWKV4G.mjs';
-import './chunk-E5F2LQCX.mjs';
-import { parseDuration, formatDuration, isTraceEvent, getTraceFilePath } from './chunk-WRE4CNPF.mjs';
+import './chunk-T5MASTIW.mjs';
+import { parseDuration, isTraceEvent, getTraceFilePath } from './chunk-WRE4CNPF.mjs';
 export { DEFAULT_TRACE_DIR_NAME, FALLBACK_TRACE_DIR, MAX_NAME_LENGTH, RUNS_DIR_NAME, createRunId, createStepId, ensureTraceDir, formatDuration, formatError, formatTimestamp, getDefaultTraceDir, getTraceFilePath, isStepStatus, isStepType, isTraceEvent, parseDuration, parseTraceJsonl, truncateName, unknownTraceFormatMessage, warn } from './chunk-WRE4CNPF.mjs';
 import './chunk-P366CCL2.mjs';
 import './chunk-7TGZLWEE.mjs';
@@ -23,41 +23,13 @@ import path from 'path';
 import { access, readFile } from 'fs/promises';
 import { pathToFileURL } from 'url';
 
-// packages/core/src/trace-filter.ts
-function toLower(s) {
-  return typeof s === "string" ? s.toLowerCase() : "";
-}
-function filterTraces(traces, options) {
-  const input = [...traces];
-  let out = input.filter((t) => {
-    if (options.status && t.status !== options.status) return false;
-    if (options.name) {
-      const q = options.name.toLowerCase();
-      const hay = `${toLower(t.name)} ${toLower(t.runId)}`;
-      if (!hay.includes(q)) return false;
-    }
-    if (options.since) {
-      const windowMs = parseDuration(options.since);
-      const cutoff = Date.now() - windowMs;
-      const started = typeof t.startedAt === "number" ? t.startedAt : void 0;
-      const basis = started ?? t.createdAt.getTime();
-      if (!Number.isFinite(basis) || basis < cutoff) return false;
-    }
-    return true;
-  });
-  out.sort((a, b) => {
-    const aTime = (typeof a.startedAt === "number" ? a.startedAt : void 0) ?? a.createdAt.getTime();
-    const bTime = (typeof b.startedAt === "number" ? b.startedAt : void 0) ?? b.createdAt.getTime();
-    return bTime - aTime;
-  });
-  if (typeof options.limit === "number" && Number.isFinite(options.limit)) {
-    const n = Math.max(0, Math.floor(options.limit));
-    out = out.slice(0, n);
-  }
-  return out;
-}
-
 // packages/core/src/explain.ts
+function toDisplayStatus(status) {
+  if (status === "ok") return "success";
+  if (status === "error") return "error";
+  if (status === "running") return "running";
+  return "unknown";
+}
 function flatten(nodes, out = []) {
   for (const node of nodes) {
     out.push({ node, index: out.length + 1 });
@@ -115,7 +87,7 @@ function buildFacts(run, redactor) {
   const facts = [
     fact("run.id", "Run id", run.runId, redactor),
     fact("run.name", "Run name", run.name ?? run.runId, redactor),
-    fact("run.status", "Run status", run.status ?? "unknown", redactor),
+    fact("run.status", "Run status", toDisplayStatus(run.status), redactor),
     fact("run.totalEvents", "Total events", run.metadata.totalEvents, redactor),
     fact("run.stepCount", "Top-level step count", run.children.length, redactor),
     fact("run.nodeCount", "Total node count", nodes.length, redactor),
@@ -191,7 +163,7 @@ function buildLocalExplanation(run, options = {}) {
     mode,
     runId: String(redactValue(redactor, "runId", run.runId)),
     ...run.name !== void 0 ? { name: String(redactValue(redactor, "name", run.name)) } : {},
-    ...run.status !== void 0 ? { status: run.status } : {},
+    ...run.status !== void 0 ? { status: toDisplayStatus(run.status) } : {},
     redactionProfile,
     facts,
     inferences: mode === "dry-run" ? [] : buildInferences(run, facts),
@@ -202,199 +174,6 @@ function buildLocalExplanation(run, options = {}) {
   };
 }
 
-// packages/core/src/stats.ts
-function percentile(sorted, p) {
-  if (sorted.length === 0) return void 0;
-  const idx = Math.min(
-    sorted.length - 1,
-    Math.max(0, Math.ceil(p / 100 * sorted.length) - 1)
-  );
-  return sorted[idx];
-}
-async function readRunStartedMetadata(filePath) {
-  try {
-    const events = await readTraceEventsFromFile(filePath);
-    for (const event of events) {
-      if (event.event !== "run_started") continue;
-      const rs = event;
-      if (rs.metadata && typeof rs.metadata === "object") {
-        return rs.metadata;
-      }
-      return void 0;
-    }
-  } catch {
-  }
-  return void 0;
-}
-function metaMatchesCorrelation(metadata, correlationId, groupId) {
-  if (correlationId) {
-    const v = metadata?.correlationId;
-    if (typeof v !== "string" || v !== correlationId) return false;
-  }
-  if (groupId) {
-    const v = metadata?.groupId;
-    if (typeof v !== "string" || v !== groupId) return false;
-  }
-  return true;
-}
-async function buildTraceStats(metas, options) {
-  let filtered = filterTraces(metas, { since: options.since });
-  if (options.correlationId || options.groupId) {
-    const next = [];
-    for (const m of filtered) {
-      const md = await readRunStartedMetadata(m.filePath);
-      if (metaMatchesCorrelation(md, options.correlationId, options.groupId)) {
-        next.push(m);
-      }
-    }
-    filtered = next;
-  }
-  let successCount = 0;
-  let errorCount = 0;
-  let runningCount = 0;
-  let unknownCount = 0;
-  const durations = [];
-  let totalSteps = 0;
-  let totalLlmSteps = 0;
-  let totalToolSteps = 0;
-  let totalErrorSteps = 0;
-  const slowestRuns = [];
-  const slowestSteps = [];
-  for (const m of filtered) {
-    if (m.status === "success") successCount += 1;
-    else if (m.status === "error") errorCount += 1;
-    else if (m.status === "running") runningCount += 1;
-    else unknownCount += 1;
-    if (typeof m.durationMs === "number" && Number.isFinite(m.durationMs) && m.durationMs >= 0) {
-      durations.push(m.durationMs);
-      slowestRuns.push({
-        runId: m.runId,
-        name: m.name,
-        durationMs: m.durationMs,
-        status: m.status
-      });
-    }
-    try {
-      const events = await readTraceEventsFromFile(m.filePath);
-      if (events.length === 0) continue;
-      const summary = buildRunSummary(events);
-      totalSteps += summary.totalSteps;
-      totalLlmSteps += summary.llmSteps;
-      totalToolSteps += summary.toolSteps;
-      totalErrorSteps += summary.errorSteps;
-      const steps = collectCompletedSteps(events, m.runId);
-      for (const s of steps) {
-        slowestSteps.push(s);
-      }
-    } catch {
-    }
-  }
-  slowestRuns.sort((a, b) => (b.durationMs ?? 0) - (a.durationMs ?? 0));
-  slowestSteps.sort((a, b) => b.durationMs - a.durationMs);
-  const runLimit = options.slowRunLimit ?? 5;
-  const stepLimit = options.slowStepLimit ?? 5;
-  const sortedDur = [...durations].sort((a, b) => a - b);
-  const totalRuns = filtered.length;
-  const errorRate = totalRuns > 0 ? errorCount / totalRuns : 0;
-  const sumDur = durations.reduce((a, b) => a + b, 0);
-  return {
-    traceDir: options.traceDir,
-    ...options.since ? { since: options.since } : {},
-    ...options.correlationId ? { correlationId: options.correlationId } : {},
-    ...options.groupId ? { groupId: options.groupId } : {},
-    totalRuns,
-    successCount,
-    errorCount,
-    runningCount,
-    unknownCount,
-    errorRate,
-    duration: {
-      ...sortedDur.length > 0 ? {
-        minMs: sortedDur[0],
-        maxMs: sortedDur[sortedDur.length - 1],
-        avgMs: sumDur / sortedDur.length,
-        p50Ms: percentile(sortedDur, 50),
-        p95Ms: percentile(sortedDur, 95)
-      } : {}
-    },
-    totalSteps,
-    avgStepsPerRun: totalRuns > 0 ? totalSteps / totalRuns : 0,
-    totalLlmSteps,
-    totalToolSteps,
-    totalErrorSteps,
-    slowestRuns: slowestRuns.slice(0, runLimit),
-    slowestSteps: slowestSteps.slice(0, stepLimit)
-  };
-}
-function collectCompletedSteps(events, runId) {
-  const started = /* @__PURE__ */ new Map();
-  const out = [];
-  for (const e of events) {
-    if (e.event === "step_started") {
-      const s = e;
-      started.set(s.stepId, { name: s.name, type: s.type });
-    }
-    if (e.event === "step_completed") {
-      const c = e;
-      if (c.status !== "success" && c.status !== "error") continue;
-      if (typeof c.durationMs !== "number" || !Number.isFinite(c.durationMs)) {
-        continue;
-      }
-      const meta = started.get(c.stepId);
-      out.push({
-        runId,
-        stepName: meta?.name ?? c.stepId,
-        stepType: meta?.type ?? "logic",
-        durationMs: c.durationMs
-      });
-    }
-  }
-  return out;
-}
-function renderTraceStats(stats) {
-  const lines = [];
-  lines.push("Trace stats (local)");
-  lines.push(`Directory: ${stats.traceDir}`);
-  if (stats.since) lines.push(`Since: ${stats.since}`);
-  if (stats.correlationId) lines.push(`Correlation ID: ${stats.correlationId}`);
-  if (stats.groupId) lines.push(`Group ID: ${stats.groupId}`);
-  lines.push("");
-  lines.push(`Runs: ${stats.totalRuns}`);
-  lines.push(
-    `  success: ${stats.successCount}  error: ${stats.errorCount}  running: ${stats.runningCount}  unknown: ${stats.unknownCount}`
-  );
-  lines.push(`Error rate: ${(stats.errorRate * 100).toFixed(1)}%`);
-  if (stats.duration.avgMs !== void 0) {
-    lines.push(
-      `Duration: min ${formatDuration(stats.duration.minMs ?? 0)} | avg ${formatDuration(stats.duration.avgMs)} | p50 ${formatDuration(stats.duration.p50Ms ?? 0)} | p95 ${formatDuration(stats.duration.p95Ms ?? 0)} | max ${formatDuration(stats.duration.maxMs ?? 0)}`
-    );
-  }
-  lines.push("");
-  lines.push(`Steps: ${stats.totalSteps} (avg ${stats.avgStepsPerRun.toFixed(1)} per run)`);
-  lines.push(
-    `  LLM: ${stats.totalLlmSteps}  tool: ${stats.totalToolSteps}  errors: ${stats.totalErrorSteps}`
-  );
-  if (stats.slowestRuns.length > 0) {
-    lines.push("");
-    lines.push("Slowest runs:");
-    for (const r of stats.slowestRuns) {
-      lines.push(
-        `  ${r.runId} | ${r.name ?? "-"} | ${formatDuration(r.durationMs ?? 0)} | ${r.status}`
-      );
-    }
-  }
-  if (stats.slowestSteps.length > 0) {
-    lines.push("");
-    lines.push("Slowest steps:");
-    for (const s of stats.slowestSteps) {
-      lines.push(
-        `  ${s.runId} | ${s.stepType}:${s.stepName} | ${formatDuration(s.durationMs)}`
-      );
-    }
-  }
-  return lines.join("\n");
-}
-
 // packages/core/src/search.ts
 function parseDurationFilter(expr) {
   const raw = expr.trim();
@@ -461,6 +240,12 @@ async function searchTraces(metas, options) {
         ...sessionLabel ? { sessionId: sessionLabel } : {}
       });
     }
+    results.sort((a, b) => {
+      const ta = a.timestamp ?? 0;
+      const tb = b.timestamp ?? 0;
+      if (ta !== tb) return tb - ta;
+      return a.runId.localeCompare(b.runId);
+    });
     return results.slice(0, limit);
   }
   for (const m of filtered) {
@@ -508,7 +293,7 @@ async function searchTraces(metas, options) {
   results.sort((a, b) => {
     const ta = a.timestamp ?? 0;
     const tb = b.timestamp ?? 0;
-    if (ta !== tb) return ta - tb;
+    if (ta !== tb) return tb - ta;
     const runCmp = a.runId.localeCompare(b.runId);
     if (runCmp !== 0) return runCmp;
     return (a.stepName ?? "").localeCompare(b.stepName ?? "");
@@ -2417,7 +2202,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average duration (ms)",
           baselineAgg?.avgDurationMs,
-          candidateAgg?.avgDurationMs
+          candidateAgg?.avgDurationMs,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2427,7 +2214,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average LLM calls",
           baselineAgg?.avgLlmCallCount,
-          candidateAgg?.avgLlmCallCount
+          candidateAgg?.avgLlmCallCount,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2437,7 +2226,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average token usage",
           baselineAgg?.avgTokenUsage,
-          candidateAgg?.avgTokenUsage
+          candidateAgg?.avgTokenUsage,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2447,7 +2238,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average retries",
           baselineAgg?.avgRetryCount,
-          candidateAgg?.avgRetryCount
+          candidateAgg?.avgRetryCount,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2457,7 +2250,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Observation failure rate",
           baselineAgg?.observationFailureRate,
-          candidateAgg?.observationFailureRate
+          candidateAgg?.observationFailureRate,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2493,7 +2288,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Guardrail failures",
           baselineAgg?.avgGuardrailFailures,
-          candidateAgg?.avgGuardrailFailures
+          candidateAgg?.avgGuardrailFailures,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2503,7 +2300,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Circuit violations",
           baselineAgg?.avgCircuitViolations,
-          candidateAgg?.avgCircuitViolations
+          candidateAgg?.avgCircuitViolations,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2513,7 +2312,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Redaction warnings",
           baselineAgg?.avgRedactionWarnings,
-          candidateAgg?.avgRedactionWarnings
+          candidateAgg?.avgRedactionWarnings,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -2633,7 +2434,7 @@ async function computeCohortRunMetrics(input) {
     toolOrdering: tools.ordering
   };
 }
-function percentile2(values, p) {
+function percentile(values, p) {
   if (values.length === 0) return void 0;
   const sorted = [...values].sort((a, b) => a - b);
   const idx = Math.min(
@@ -2687,7 +2488,7 @@ function aggregateCohortMetrics(runs, groupKey, cohortLabel) {
     runCount: runs.length,
     errorRate: runs.length > 0 ? errors / runs.length : 0,
     avgDurationMs: durations.length > 0 ? durations.reduce((sum, value) => sum + value, 0) / durations.length : void 0,
-    p95DurationMs: percentile2(durations, 95),
+    p95DurationMs: percentile(durations, 95),
     avgLlmCallCount: runs.length > 0 ? runs.reduce((sum, run) => sum + run.llmCallCount, 0) / runs.length : 0,
     avgTokenUsage: tokenValues.length > 0 ? tokenValues.reduce((sum, value) => sum + value, 0) / tokenValues.length : void 0,
     avgRetryCount: runs.length > 0 ? runs.reduce((sum, run) => sum + run.retryCount, 0) / runs.length : 0,
@@ -2913,7 +2714,7 @@ function parseGateNumber(value, label, options = {}) {
 }
 
 // packages/core/src/gate/evaluate.ts
-function percentile3(values, p) {
+function percentile2(values, p) {
   if (values.length === 0) return void 0;
   const sorted = [...values].sort((a, b) => a - b);
   const idx = Math.min(
@@ -2985,7 +2786,7 @@ async function evaluateGateThresholds(runs, options) {
   }
   if (options.maxP95DurationMs !== void 0) {
     const durations = runMetrics.map((run) => run.durationMs).filter((value) => typeof value === "number");
-    const actual = percentile3(durations, 95);
+    const actual = percentile2(durations, 95);
     const ok = actual !== void 0 && actual <= options.maxP95DurationMs;
     checks.push({
       id: "maxP95Duration",
@@ -3297,6 +3098,6 @@ function renderGateReport(result, options = {}) {
   return renderGateSummaryMarkdown(result);
 }
 
-export { COHORT_METRIC_IDS, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, SESSION_WORKFLOW_KEYS, SUITE_TEMPLATE_IDS, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, assertBundlePathContained, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, bundleRunAssetRelativePath, compareCohortAggregates, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, gateHasThresholds, getSuiteTemplate, groupSessionCohorts, isAgentInspectTrace, listSuiteTemplates, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderSuiteReport, renderSuiteReportMarkdown, renderTraceStats, resolveBundleRunIds, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveSuiteTemplate, runGate, runSuite, sanitizeBundleRunId, searchTraces, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, validateSuiteConfig };
+export { COHORT_METRIC_IDS, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, SESSION_WORKFLOW_KEYS, SUITE_TEMPLATE_IDS, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, assertBundlePathContained, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildSessionIndex, bundleFailsOnSafety, bundleRunAssetRelativePath, compareCohortAggregates, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, extractSessionWorkflowMetadata, filterMetasBySessionScope, gateHasThresholds, getSuiteTemplate, groupSessionCohorts, isAgentInspectTrace, listSuiteTemplates, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderSuiteReport, renderSuiteReportMarkdown, resolveBundleRunIds, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveSuiteTemplate, runGate, runSuite, sanitizeBundleRunId, searchTraces, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, validateSuiteConfig };
 //# sourceMappingURL=advanced.mjs.map
 //# sourceMappingURL=advanced.mjs.map