agent-inspect 6.7.3 → 6.7.4

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## 6.7.4
+
+### Patch Changes
+
+- ab2ad83: Real-integration blocker patch: standalone LangGraph-shaped callback runs complete via active lifecycle; CLI shorthand check flags auto-select their rules; human tool display names; shared step labels and newest-first search; cross-command run-status golden; synthetic LangGraph fixtures; publish prior RUN-lifecycle and stats label fixes.
+
 ## 6.7.3
 
 ### Patch Changes
@@ -8,6 +14,15 @@
 
 ## Unreleased
 
+### Documentation
+
+- Activate the Stability and Focus roadmap (v6.7.3 → v6.12 → conditional v7): operational source-of-truth reconciliation, baseline audit, and release-train plans. No runtime or schema change in this docs activation.
+
+### Patch candidates (unpublished on main; intended for 6.7.4)
+
+- Completed-run status derived from terminal RUN lifecycle; explain uses shared user-facing status vocabulary (`8e525f1`).
+- Stats no longer double-prefixes already-typed step names (`tool:tool:` / `llm:llm:`) (`ee49d4c`).
+
 ## 6.7.2
 
 ### Patch Changes

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-inspect",
-  "version": "6.7.3",
+  "version": "6.7.4",
   "license": "MIT",
   "type": "module",
   "description": "Debug, regression-test, and safely share TypeScript AI-agent behavior locally — no account, no default upload, metadata-only by default",

package/packages/cli/dist/{chunk-EMYDJREA.mjs → chunk-FVEQ7JFF.mjs}

@@ -1904,6 +1904,202 @@ function filterTraces(traces, options) {
   return out;
 }
 
+// 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 formatStepLabel(stepType, stepName) {
+  return stepName.startsWith(`${stepType}:`) ? stepName : `${stepType}:${stepName}`;
+}
+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 ${formatDuration2(stats.duration.minMs ?? 0)} | avg ${formatDuration2(stats.duration.avgMs)} | p50 ${formatDuration2(stats.duration.p50Ms ?? 0)} | p95 ${formatDuration2(stats.duration.p95Ms ?? 0)} | max ${formatDuration2(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 ?? "-"} | ${formatDuration2(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} | ${formatStepLabel(s.stepType, s.stepName)} | ${formatDuration2(s.durationMs)}`
+      );
+    }
+  }
+  return lines.join("\n");
+}
+
 // packages/core/src/timeline.ts
 function finite(n) {
   return typeof n === "number" && Number.isFinite(n);
@@ -2061,7 +2257,7 @@ function renderTimeline(timeline, options = {}) {
     const dur = e.durationMs !== void 0 ? formatDuration2(e.durationMs) : "-";
     const err = e.isError ? " error" : "";
     const off = formatDuration2(e.offsetMs);
-    let line = `${prefix}+${off} ${typeTag}:${e.name} (${dur})${err}`;
+    let line = `${prefix}+${off} ${formatStepLabel(typeTag, e.name)} (${dur})${err}`;
     if (e.streaming?.chunkCount !== void 0) {
       line += ` chunks=${e.streaming.chunkCount}`;
     }
@@ -2213,6 +2409,12 @@ function renderRunWhat(summary, options = {}) {
 }
 
 // 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 });
@@ -2270,7 +2472,7 @@ 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),
@@ -2346,7 +2548,7 @@ 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),
@@ -2357,199 +2559,6 @@ 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 ${formatDuration2(stats.duration.minMs ?? 0)} | avg ${formatDuration2(stats.duration.avgMs)} | p50 ${formatDuration2(stats.duration.p50Ms ?? 0)} | p95 ${formatDuration2(stats.duration.p95Ms ?? 0)} | max ${formatDuration2(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 ?? "-"} | ${formatDuration2(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} | ${formatDuration2(s.durationMs)}`
-      );
-    }
-  }
-  return lines.join("\n");
-}
-
 // packages/core/src/outcomes/types.ts
 var OBSERVED_OUTCOME_STATUSES = [
   "passed",
@@ -2752,6 +2761,12 @@ 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) {
@@ -2799,7 +2814,7 @@ 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 ?? "");
@@ -5814,13 +5829,24 @@ function inc(map, key) {
   map[key] = (map[key] ?? 0) + 1;
 }
 function computeRunStatus(events) {
+  let runTerminal;
+  let sawRunEvent = false;
+  for (const e of events) {
+    if (e.kind !== "RUN") continue;
+    sawRunEvent = true;
+    if (e.status === "ok" || e.status === "error") {
+      runTerminal = e.status;
+    }
+  }
+  if (sawRunEvent) {
+    return runTerminal ?? "running";
+  }
   let hasRunning = false;
   for (const e of events) {
     if (e.status === "error") return "error";
     if (e.status === "running") hasRunning = true;
   }
-  if (hasRunning) return "running";
-  return "ok";
+  return hasRunning ? "running" : "ok";
 }
 var TreeBuilder = class {
   constructor(options) {
@@ -9061,5 +9087,5 @@ function renderGateReport(result, options = {}) {
 }
 
 export { COHORT_METRIC_IDS, DEFAULT_SUITE_ARTIFACTS_DIR, Redactor, TraceDirectory, TraceReadError, TreeBuilder, __commonJS, __require, __toESM, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, applyProfileMetadataCaps, assertBundlePathContained, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, bundleRunAssetRelativePath, compactAttributes3 as compactAttributes, createBaselineRegressionRule, createLlmUsageRule, createMaxStepDurationRule, createObservedOutcomeRule, createRequireCompletedRule, createRunDepthRule, createRunDurationRule, createRunStatusRule, createSafetyOversizedAttributeRule, createSafetyRawContentRule, createSafetyRedactionRule, createSafetySecretPatternRule, createStallDetectionRule, createStructureCycleRule, createStructureOrphanRule, createStructureParallelWidthRule, createStructureRelationshipRule, createToolUsageRule, defaultBundleOutputPath, defaultSuiteConfigTemplate, enrichSessionRunRecord, escapeHtml, escapeMarkdown, extractMetadata, extractOutcomesFromTraceEvents, filterMetasBySessionScope, filterTraces, flattenTree, formatDuration2 as formatDuration, formatTimestamp, gateHasThresholds, getIndent, getTraceFilePath, isAgentInspectTrace, isPersistedInspectEvent, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, nanoid, normalizeBundleOutputPath, openTrace, parseCohortMetricList, parseDuration, parseDurationFilter, parseGateList, parseTraceJsonl, persistedInspectEventsToTraceEvents, renderActivitySummaryHuman, renderCohortReport, renderErrorLine, renderGateReport, renderObservedOutcomesHtml, renderObservedOutcomesMarkdown, renderRunWhat, renderStepLine, renderSuiteReport, renderTimeline, renderTraceStats, resolveBundleRunIds, resolveRedactionProfile, resolveSuiteTemplate, resolveTraceDir, runGate, runSuite, runTraceChecks, safeString, sanitizeBundleRunId, searchTraces, source_default, stableJson, summarizeObservedOutcomes, traceEventToPersistedInspectEvent, truncateName, truncateStringForProfile, validateEvent, validateSuiteConfig, zeroKinds };
-//# sourceMappingURL=chunk-EMYDJREA.mjs.map
-//# sourceMappingURL=chunk-EMYDJREA.mjs.map
+//# sourceMappingURL=chunk-FVEQ7JFF.mjs.map
+//# sourceMappingURL=chunk-FVEQ7JFF.mjs.map