agent-inspect 6.31.11 → 6.31.13

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # Changelog
 
+## 6.31.13
+
+### Patch Changes
+
+- be67c1b: Stability C04–C06: prefer explicit RUN start/end/duration in tree summaries; exclude OUTCOME from execution-step counts and slowest ranking; bounded recursive OTLP AnyValue/int64/uint64 producer validation.
+
+## 6.31.12
+
+### Patch Changes
+
+- 31d187e: Stability C00–C03: reconcile release-state tables to 6.31.11; Promptfoo recipe/reviewer-kit verifier freshness and exact matrix identity; strict nested suite config (no string/typo coercion); OpenAI Node recipe runs SDK create inside inspector.llm with RequestOptions and distinct HTTP vs completion IDs.
+
 ## 6.31.11
 
 ### Patch Changes

package/README.md

@@ -213,7 +213,7 @@ The root package is enough for custom capture, the CLI, checks, and Evidence wor
 
 ## Status and documentation
 
-**Current published baseline:** **6.31.11** · persisted schema `1.0` · Node.js `>=20` · MIT.
+**Current published baseline:** **6.31.13** · persisted schema `1.0` · Node.js `>=20` · MIT.
 
 Legacy v0.1 and v0.2 traces remain readable. Check the npm badge and [changelog](CHANGELOG.md) for the current published version.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-inspect",
-  "version": "6.31.11",
+  "version": "6.31.13",
   "license": "MIT",
   "type": "module",
   "description": "Local evidence debugger and trajectory-test toolkit for TypeScript AI agents — execution trees, TraceContract checks, Evidence v2, and read-only MCP",

package/packages/cli/dist/{chunk-MQYHJL3W.mjs → chunk-ZAYVAHIH.mjs}

@@ -3002,12 +3002,14 @@ function topKinds(run) {
 function countErrorNodes(nodes) {
   return nodes.filter((entry) => entry.node.event.status === "error").length;
 }
+function isExecutionStepKind(kind) {
+  return kind !== "RUN" && kind !== "OUTCOME";
+}
 function slowestNode(nodes) {
   return nodes.filter(
     (entry) => entry.node.event.durationMs !== void 0 && // The RUN boundary spans the whole run, so its duration always dominates
-    // and shadows every real step. Rank actual work nodes only, matching the
-    // stats slowest-step ranking, which never counts the run envelope.
-    entry.node.event.kind !== "RUN"
+    // and shadows every real step. OUTCOME is a linked result, not execution.
+    isExecutionStepKind(entry.node.event.kind)
   ).sort((a, b) => {
     const delta = (b.node.event.durationMs ?? 0) - (a.node.event.durationMs ?? 0);
     return delta !== 0 ? delta : a.index - b.index;
@@ -3060,6 +3062,12 @@ function buildFacts(run, redactor, events) {
     const scoped = events.filter((event) => event.runId === run.runId);
     const { logicalEvents: logical } = projectLogicalEvents(scoped);
     const logicalSteps = logical.filter((event) => event.kind !== "RUN");
+    const executionSteps = logical.filter((event) => isExecutionStepKind(event.kind));
+    const executionByKind = {
+      LLM: executionSteps.filter((event) => event.kind === "LLM").length,
+      TOOL: executionSteps.filter((event) => event.kind === "TOOL").length,
+      LOGIC: executionSteps.filter((event) => event.kind === "LOGIC").length
+    };
     facts.push(
       fact("run.rawEventCount", "Raw persisted event count", scoped.length, redactor),
       fact(
@@ -3073,9 +3081,21 @@ function buildFacts(run, redactor, events) {
         "Logical lifecycle event count (including RUN)",
         logical.length,
         redactor
+      ),
+      fact(
+        "run.executionStepCount",
+        "Execution step count (excludes RUN and OUTCOME)",
+        executionSteps.length,
+        redactor
+      ),
+      fact(
+        "run.executionStepKinds",
+        "Execution step kind counts",
+        executionByKind,
+        redactor
       )
     );
-    const slowestLogical = [...logicalSteps].filter((event) => typeof event.durationMs === "number").sort((a, b) => {
+    const slowestLogical = [...executionSteps].filter((event) => typeof event.durationMs === "number").sort((a, b) => {
       const delta = (b.durationMs ?? 0) - (a.durationMs ?? 0);
       return delta !== 0 ? delta : a.eventId.localeCompare(b.eventId);
     })[0];
@@ -11324,25 +11344,86 @@ function asStringArray(value, label) {
 function asPositiveNumber(value, label) {
   if (value === void 0) return void 0;
   if (typeof value !== "number" || !Number.isFinite(value) || value < 0) {
-    throw new Error(`${label} must be a non-negative number.`);
+    throw new Error(`${label} must be a non-negative finite number.`);
   }
   return value;
 }
-function asExpectConfig(value, label) {
+function asBoolean(value, label) {
   if (value === void 0) return void 0;
+  if (typeof value !== "boolean") {
+    throw new Error(`${label} must be a boolean (not a string or other type).`);
+  }
+  return value;
+}
+function assertObject(value, label) {
   if (typeof value !== "object" || value === null || Array.isArray(value)) {
     throw new Error(`${label} must be an object.`);
   }
-  const raw = value;
+  return value;
+}
+function assertKnownKeys(raw, allowed, label) {
+  const allowedSet = new Set(allowed);
+  for (const key of Object.keys(raw)) {
+    if (!allowedSet.has(key)) {
+      throw new Error(`${label} has unknown key "${key}".`);
+    }
+  }
+}
+var CASE_KEYS = [
+  "id",
+  "trace",
+  "runId",
+  "input",
+  "requireTools",
+  "forbidTools",
+  "maxDurationMs",
+  "expectedObservations",
+  "expect"
+];
+var EXPECT_KEYS = [
+  "checkStatus",
+  "findingRuleIds",
+  "rejectUnexpectedFindings",
+  "minAssertions"
+];
+var CHECKS_KEYS = ["select", "run", "tool", "llm"];
+var CHECKS_RUN_KEYS = ["maxDurationMs", "maxDepth"];
+var CHECKS_TOOL_KEYS = ["required", "forbidden"];
+var CHECKS_LLM_KEYS = ["allowedModels", "maxTotalTokens"];
+var EVAL_KEYS = [
+  "requireSuccess",
+  "requiredTools",
+  "forbiddenTools",
+  "maxDurationMs",
+  "maxDepth",
+  "maxRetries",
+  "maxTotalTokens"
+];
+var ROOT_KEYS = [
+  "name",
+  "traces",
+  "cases",
+  "checks",
+  "eval",
+  "redactionProfile",
+  "artifacts",
+  "baseline",
+  "candidate"
+];
+function asExpectConfig(value, label) {
+  if (value === void 0) return void 0;
+  const raw = assertObject(value, label);
+  assertKnownKeys(raw, EXPECT_KEYS, label);
   const checkStatus = raw.checkStatus;
   if (checkStatus !== "pass" && checkStatus !== "fail" && checkStatus !== "error") {
     throw new Error(`${label}.checkStatus must be "pass", "fail", or "error".`);
   }
   const findingRuleIds = asStringArray(raw.findingRuleIds, `${label}.findingRuleIds`);
   const minAssertions = asPositiveNumber(raw.minAssertions, `${label}.minAssertions`);
-  const rejectUnexpectedFindings = raw.rejectUnexpectedFindings === void 0 ? void 0 : typeof raw.rejectUnexpectedFindings === "boolean" ? raw.rejectUnexpectedFindings : (() => {
-    throw new Error(`${label}.rejectUnexpectedFindings must be a boolean.`);
-  })();
+  const rejectUnexpectedFindings = asBoolean(
+    raw.rejectUnexpectedFindings,
+    `${label}.rejectUnexpectedFindings`
+  );
   return {
     checkStatus,
     ...findingRuleIds !== void 0 ? { findingRuleIds } : {},
@@ -11350,6 +11431,56 @@ function asExpectConfig(value, label) {
     ...minAssertions !== void 0 ? { minAssertions } : {}
   };
 }
+function asChecksConfig(value, label) {
+  const raw = assertObject(value, label);
+  assertKnownKeys(raw, CHECKS_KEYS, label);
+  const checks = {};
+  const select = asStringArray(raw.select, `${label}.select`);
+  if (select !== void 0) checks.select = select;
+  if (raw.run !== void 0) {
+    const run = assertObject(raw.run, `${label}.run`);
+    assertKnownKeys(run, CHECKS_RUN_KEYS, `${label}.run`);
+    checks.run = {
+      ...asPositiveNumber(run.maxDurationMs, `${label}.run.maxDurationMs`) !== void 0 ? { maxDurationMs: asPositiveNumber(run.maxDurationMs, `${label}.run.maxDurationMs`) } : {},
+      ...asPositiveNumber(run.maxDepth, `${label}.run.maxDepth`) !== void 0 ? { maxDepth: asPositiveNumber(run.maxDepth, `${label}.run.maxDepth`) } : {}
+    };
+  }
+  if (raw.tool !== void 0) {
+    const tool = assertObject(raw.tool, `${label}.tool`);
+    assertKnownKeys(tool, CHECKS_TOOL_KEYS, `${label}.tool`);
+    checks.tool = {
+      ...asStringArray(tool.required, `${label}.tool.required`) !== void 0 ? { required: asStringArray(tool.required, `${label}.tool.required`) } : {},
+      ...asStringArray(tool.forbidden, `${label}.tool.forbidden`) !== void 0 ? { forbidden: asStringArray(tool.forbidden, `${label}.tool.forbidden`) } : {}
+    };
+  }
+  if (raw.llm !== void 0) {
+    const llm = assertObject(raw.llm, `${label}.llm`);
+    assertKnownKeys(llm, CHECKS_LLM_KEYS, `${label}.llm`);
+    checks.llm = {
+      ...asStringArray(llm.allowedModels, `${label}.llm.allowedModels`) !== void 0 ? { allowedModels: asStringArray(llm.allowedModels, `${label}.llm.allowedModels`) } : {},
+      ...asPositiveNumber(llm.maxTotalTokens, `${label}.llm.maxTotalTokens`) !== void 0 ? {
+        maxTotalTokens: asPositiveNumber(
+          llm.maxTotalTokens,
+          `${label}.llm.maxTotalTokens`
+        )
+      } : {}
+    };
+  }
+  return checks;
+}
+function asEvalConfig(value, label) {
+  const raw = assertObject(value, label);
+  assertKnownKeys(raw, EVAL_KEYS, label);
+  return {
+    ...asBoolean(raw.requireSuccess, `${label}.requireSuccess`) !== void 0 ? { requireSuccess: asBoolean(raw.requireSuccess, `${label}.requireSuccess`) } : {},
+    ...asStringArray(raw.requiredTools, `${label}.requiredTools`) !== void 0 ? { requiredTools: asStringArray(raw.requiredTools, `${label}.requiredTools`) } : {},
+    ...asStringArray(raw.forbiddenTools, `${label}.forbiddenTools`) !== void 0 ? { forbiddenTools: asStringArray(raw.forbiddenTools, `${label}.forbiddenTools`) } : {},
+    ...asPositiveNumber(raw.maxDurationMs, `${label}.maxDurationMs`) !== void 0 ? { maxDurationMs: asPositiveNumber(raw.maxDurationMs, `${label}.maxDurationMs`) } : {},
+    ...asPositiveNumber(raw.maxDepth, `${label}.maxDepth`) !== void 0 ? { maxDepth: asPositiveNumber(raw.maxDepth, `${label}.maxDepth`) } : {},
+    ...asPositiveNumber(raw.maxRetries, `${label}.maxRetries`) !== void 0 ? { maxRetries: asPositiveNumber(raw.maxRetries, `${label}.maxRetries`) } : {},
+    ...asPositiveNumber(raw.maxTotalTokens, `${label}.maxTotalTokens`) !== void 0 ? { maxTotalTokens: asPositiveNumber(raw.maxTotalTokens, `${label}.maxTotalTokens`) } : {}
+  };
+}
 function validateCaseConfig(value, index) {
   const diagnostics = [];
   if (typeof value !== "object" || value === null || Array.isArray(value)) {
@@ -11360,6 +11491,7 @@ function validateCaseConfig(value, index) {
   }
   const raw = value;
   try {
+    assertKnownKeys(raw, CASE_KEYS, `cases[${index}]`);
     const id = asString(raw.id, `cases[${index}].id`);
     if (id === void 0) {
       diagnostics.push(
@@ -11408,6 +11540,7 @@ function normalizeSuiteConfig(value) {
     throw new Error("Suite config must export an object.");
   }
   const raw = value;
+  assertKnownKeys(raw, ROOT_KEYS, "suite config");
   const name = asString(raw.name, "name");
   const traces = asString(raw.traces, "traces");
   if (name === void 0) throw new Error("name is required.");
@@ -11436,25 +11569,15 @@ function normalizeSuiteConfig(value) {
   }
   const config = { name, traces, cases };
   if (raw.checks !== void 0) {
-    if (typeof raw.checks !== "object" || Array.isArray(raw.checks)) {
-      throw new Error("checks must be an object.");
-    }
-    config.checks = raw.checks;
+    config.checks = asChecksConfig(raw.checks, "checks");
   }
   if (raw.eval !== void 0) {
-    if (typeof raw.eval !== "object" || Array.isArray(raw.eval)) {
-      throw new Error("eval must be an object.");
-    }
-    config.eval = raw.eval;
+    config.eval = asEvalConfig(raw.eval, "eval");
   }
   if (raw.artifacts !== void 0) {
-    if (typeof raw.artifacts !== "object" || Array.isArray(raw.artifacts)) {
-      throw new Error("artifacts must be an object.");
-    }
-    const outputDir = asString(
-      raw.artifacts.outputDir,
-      "artifacts.outputDir"
-    );
+    const artifacts = assertObject(raw.artifacts, "artifacts");
+    assertKnownKeys(artifacts, ["outputDir"], "artifacts");
+    const outputDir = asString(artifacts.outputDir, "artifacts.outputDir");
     config.artifacts = outputDir !== void 0 ? { outputDir } : {};
   }
   if (raw.baseline !== void 0) {
@@ -12007,18 +12130,16 @@ var TreeBuilder = class {
         inc(confidenceBreakdown, e.confidence);
         kinds[e.kind] = (kinds[e.kind] ?? 0) + 1;
       }
-      const startedAt = sorted.length > 0 ? sorted[0].timestamp : void 0;
-      const endedAt = sorted.length > 0 ? sorted[sorted.length - 1].timestamp : void 0;
-      const status = computeRunStatus(sorted);
-      const durationMs = startedAt !== void 0 && endedAt !== void 0 && Number.isFinite(startedAt) && Number.isFinite(endedAt) && endedAt >= startedAt && status !== "running" ? endedAt - startedAt : void 0;
       const name = sorted.find((e) => e.kind === "RUN")?.name;
+      const status = computeRunStatus(sorted);
+      const timing = resolveRunTiming(sorted, status);
       out.push({
         runId,
         name,
         status,
-        startedAt,
-        endedAt: status === "running" ? void 0 : endedAt,
-        durationMs,
+        startedAt: timing.startedAt,
+        endedAt: timing.endedAt,
+        durationMs: timing.durationMs,
         children: roots,
         metadata: {
           totalEvents: sorted.length,
@@ -12032,6 +12153,57 @@ var TreeBuilder = class {
     return out;
   }
 };
+function resolveRunTiming(sorted, status) {
+  let startedAtMs;
+  let endedAtMs;
+  let explicitDurationMs;
+  for (const event of sorted) {
+    if (event.kind !== "RUN") continue;
+    const attrs = event.attributes ?? {};
+    if (typeof attrs.startedAtMs === "number" && Number.isFinite(attrs.startedAtMs)) {
+      startedAtMs = startedAtMs === void 0 ? attrs.startedAtMs : Math.min(startedAtMs, attrs.startedAtMs);
+    }
+    if (typeof attrs.endedAtMs === "number" && Number.isFinite(attrs.endedAtMs)) {
+      endedAtMs = endedAtMs === void 0 ? attrs.endedAtMs : Math.max(endedAtMs, attrs.endedAtMs);
+    }
+    if ((event.status === "ok" || event.status === "error") && typeof event.durationMs === "number" && Number.isFinite(event.durationMs) && event.durationMs >= 0) {
+      explicitDurationMs = event.durationMs;
+      if (endedAtMs === void 0 && startedAtMs !== void 0) {
+        endedAtMs = startedAtMs + event.durationMs;
+      } else if (endedAtMs === void 0 && Number.isFinite(event.timestamp)) {
+        endedAtMs = event.timestamp + event.durationMs;
+        if (startedAtMs === void 0) {
+          startedAtMs = event.timestamp;
+        }
+      }
+    }
+    if (startedAtMs === void 0 && event.status === "running" && Number.isFinite(event.timestamp)) {
+      startedAtMs = event.timestamp;
+    }
+  }
+  if (startedAtMs !== void 0 && endedAtMs !== void 0 && Number.isFinite(startedAtMs) && Number.isFinite(endedAtMs) && endedAtMs >= startedAtMs && status !== "running") {
+    return {
+      startedAt: startedAtMs,
+      endedAt: endedAtMs,
+      durationMs: endedAtMs - startedAtMs
+    };
+  }
+  if (startedAtMs !== void 0 && explicitDurationMs !== void 0 && status !== "running") {
+    return {
+      startedAt: startedAtMs,
+      endedAt: startedAtMs + explicitDurationMs,
+      durationMs: explicitDurationMs
+    };
+  }
+  const startedAt = sorted.length > 0 ? sorted[0].timestamp : void 0;
+  const endedAt = sorted.length > 0 ? sorted[sorted.length - 1].timestamp : void 0;
+  const durationMs = startedAt !== void 0 && endedAt !== void 0 && Number.isFinite(startedAt) && Number.isFinite(endedAt) && endedAt >= startedAt && status !== "running" ? endedAt - startedAt : void 0;
+  return {
+    startedAt,
+    endedAt: status === "running" ? void 0 : endedAt,
+    durationMs
+  };
+}
 
 // packages/core/src/persisted/to-inspect-event.ts
 function compactAttributes3(entries) {
@@ -15375,5 +15547,5 @@ function renderGateReport(result, options = {}) {
 }
 
 export { ATTRIBUTION_CONFIDENCES, COHORT_METRIC_IDS, DEFAULT_REDACT_KEYS, DEFAULT_SUITE_ARTIFACTS_DIR, EVIDENCE_FORMAT_VERSION, EVIDENCE_HTML_FILENAME, EVIDENCE_MANIFEST_FILENAME, Redactor, TraceDirectory, TraceReadError, TreeBuilder, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, applyProfileMetadataCaps, assertBundlePathContained, assertEvidenceRelativePath, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildEvidenceCausalFailureViewHtml, buildEvidenceCiPackage, buildEvidenceCircuitViewHtml, buildEvidenceContractsViewHtml, buildEvidenceDiffViewHtml, buildEvidenceHtmlShell, buildEvidenceManifest, buildEvidenceOutcomesViewHtml, buildEvidenceProvenanceViewHtml, buildEvidenceSafetyViewHtml, buildEvidenceTimelineViewHtml, buildEvidenceToolsLlmViewHtml, buildEvidenceTreeViewHtml, buildLocalExplanation, buildPlaceholderArtifact, buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildSessionIndex, buildTraceStats, buildZipArchive, bundleFailsOnSafety, bundleRunAssetRelativePath, collectTraceSchemaVersions, compactAttributes, createBaselineRegressionRule, createLlmUsageRule, createMaxStepDurationRule, createObservedOutcomeRule, createRequireCompletedRule, createRunDepthRule, createRunDurationRule, createRunStatusRule, createSafetyOversizedAttributeRule, createSafetyRawContentRule, createSafetyRedactionRule, createSafetySecretPatternRule, createStallDetectionRule, createStructureCycleRule, createStructureOrphanRule, createStructureParallelWidthRule, createStructureRelationshipRule, createToolUsageRule, defaultBundleOutputPath, defaultSuiteConfigTemplate, defineTraceContract, diffRuns, diffTraceEvents, enrichSessionRunRecord, escapeHtml, escapeMarkdown, evaluateTraceContract, extractMetadata, extractOutcomesFromTraceEvents, filterMetasBySessionScope, filterTraces, flattenTree, formatDuration2 as formatDuration, formatStepLabel, formatTimestamp, gateHasThresholds, getIndent, getTraceFilePath, inferEvidenceFileRole, isAgentInspectTrace, isAttributionConfidence, isCredentialSensitiveKey, isPersistedInspectEvent, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, manualTraceEventsToComparableRun, nanoid, normalizeBundleOutputPath, openTrace, parseCohortMetricList, parseDuration, parseDurationFilter, parseGateList, parseTraceJsonl, persistedInspectEventsToRunTrees, persistedInspectEventsToTraceEvents, projectLogicalEvents, renderActivitySummaryHuman, renderCohortReport, renderErrorLine, renderGateReport, renderObservedOutcomesHtml, renderObservedOutcomesMarkdown, renderRunDiff, renderRunWhat, renderStepLine, renderSuiteReport, renderTimeline, renderTraceStats, resolveBundleRunIds, resolveRedactionProfile, resolveSuiteTemplate, resolveTraceDir, runGate, runSuite, runTraceChecks, safeString, sanitizeBundleRunId, searchTraces, serializeEvidenceManifest, sha256Hex, stableJson, stringContainsHighConfidenceCredential, summarizeObservedOutcomes, summarizeSemanticParity, traceEventToPersistedInspectEvent, truncateName, truncateStringForProfile, validateEvent, validateSuiteConfig, verifyEvidenceDirectory, zeroKinds };
-//# sourceMappingURL=chunk-MQYHJL3W.mjs.map
-//# sourceMappingURL=chunk-MQYHJL3W.mjs.map
+//# sourceMappingURL=chunk-ZAYVAHIH.mjs.map
+//# sourceMappingURL=chunk-ZAYVAHIH.mjs.map