agent-inspect 6.31.32 → 6.31.34

package/packages/core/dist/{chunk-6G67ZN72.mjs → chunk-CXC7L4PY.mjs}

@@ -1,4 +1,4 @@
-import { OBSERVED_OUTCOME_METHODS, extractOutcomesFromPersistedEvents, outcomesMatchingStatus } from './chunk-FBPDF7HO.mjs';
+import { OBSERVED_OUTCOME_METHODS, extractOutcomesFromPersistedEvents, outcomesMatchingStatus } from './chunk-6F6KLGCR.mjs';
 import { DEFAULT_CREDENTIAL_SENSITIVE_KEYS, isCredentialSensitiveKey } from './chunk-VLVCLAPB.mjs';
 import { projectLogicalEvents, resolveCanonicalToolName } from './chunk-AHUZNHFI.mjs';
 import { formatProgrammaticDiagnostic } from './chunk-S6LQBLX5.mjs';
@@ -2345,10 +2345,17 @@ function contractToRules(contract) {
             (event) => event.status === "running"
           );
           if (running.length > 0) {
+            const labels = running.map((event) => {
+              if (event.kind === "TOOL") {
+                return `tool ${resolveCanonicalToolName(event)}`;
+              }
+              return `${event.kind} ${event.name}`;
+            });
+            const message = labels.length === 1 ? `Run contains incomplete running events: ${labels[0]} never completed.` : `Run contains incomplete running events: ${labels.join("; ")}.`;
             findings.push(
               contractFailFinding(
                 "contract.run.allowedStatuses",
-                "Run contains incomplete running events.",
+                message,
                 running.map((event) => ({
                   runId: event.runId,
                   eventId: event.eventId,
@@ -2357,7 +2364,10 @@ function contractToRules(contract) {
                   status: event.status
                 })),
                 "no running events",
-                running.length
+                {
+                  count: running.length,
+                  unfinished: labels
+                }
               )
             );
           }
@@ -3307,10 +3317,11 @@ var SAFE_USAGE_LEAF_KEYS = /* @__PURE__ */ new Set([
 var DEFAULT_SECRET_PATTERNS = [
   { id: "bearer-token", pattern: /Bearer\s+[A-Za-z0-9._~+/-]{12,}=*/ },
   { id: "openai-key", pattern: /sk-[A-Za-z0-9_-]{16,}/ },
-  // Stripe + Slack: keep in sync with packages/redact + credential-value-patterns (#520).
+  // Stripe + Slack: keep in sync with packages/redact + credential-value-patterns (#520/#528).
+  // Non-alnum lookarounds so underscore-adjacent forms (acct_sk_live_…, …_eu) match.
   {
     id: "stripe-key",
-    pattern: /\b(?:sk|rk)_(?:live|test)_[A-Za-z0-9]{16,}\b/
+    pattern: /(?<![A-Za-z0-9])(?:sk|rk)_(?:live|test)_[A-Za-z0-9]{16,}(?![A-Za-z0-9])/
   },
   { id: "slack-token", pattern: /\bxox[abposr]-[A-Za-z0-9-]{10,}\b/ },
   { id: "aws-access-key", pattern: /AKIA[0-9A-Z]{16}/ },
@@ -3494,6 +3505,11 @@ function compareFindings(eventById) {
     if (STATUS_RANK[a.status] !== STATUS_RANK[b.status]) {
       return STATUS_RANK[a.status] - STATUS_RANK[b.status];
     }
+    if (a.ruleId === "tool.usage") {
+      const aHasEvent = a.evidence[0]?.eventId !== void 0;
+      const bHasEvent = b.evidence[0]?.eventId !== void 0;
+      if (aHasEvent !== bHasEvent) return aHasEvent ? -1 : 1;
+    }
     const byRun = compareStrings(a.evidence[0]?.runId, b.evidence[0]?.runId);
     if (byRun !== 0) return byRun;
     const byTime = eventTimestamp(a, eventById).localeCompare(eventTimestamp(b, eventById));
@@ -3619,6 +3635,37 @@ function failFinding(ruleId, message, evidence, expected, actual, meta) {
 function toolName(event) {
   return resolveCanonicalToolName(event);
 }
+function formatToolNameForMessage(name) {
+  const details = [];
+  for (let index = 0; index < name.length; ) {
+    const codePoint = name.codePointAt(index);
+    const width = codePoint > 65535 ? 2 : 1;
+    if (codePoint > 127) {
+      details.push(
+        `U+${codePoint.toString(16).toUpperCase().padStart(4, "0")} at index ${index}`
+      );
+    }
+    index += width;
+  }
+  if (details.length === 0) return name;
+  return `"${name}" (contains ${details.join("; ")})`;
+}
+function describeRunningEvent(event) {
+  if (event.kind === "TOOL") {
+    return `tool ${formatToolNameForMessage(resolveCanonicalToolName(event))}`;
+  }
+  if (event.kind === "LLM") {
+    return `llm ${stepEndpointName(event, "LLM")}`;
+  }
+  return `${event.kind} ${event.name}`;
+}
+function incompleteRunningMessage(prefix, running) {
+  const labels = running.map(describeRunningEvent);
+  if (labels.length === 1) {
+    return `${prefix}: ${labels[0]} never completed.`;
+  }
+  return `${prefix}: ${labels.join("; ")}.`;
+}
 function stepEndpointName(event, kind) {
   if (kind === "TOOL") return resolveCanonicalToolName(event);
   for (const prefix of ["llm:", "generation:", "transcription:", "speech:"]) {
@@ -3963,10 +4010,13 @@ function createRunStatusRule(options = {}) {
           findings.push(
             failFinding(
               "run.status",
-              "Run contains incomplete running events.",
+              incompleteRunningMessage("Run contains incomplete running events", running),
               running.map((event) => eventEvidence3(event)),
               "no running events",
-              running.length
+              {
+                count: running.length,
+                unfinished: running.map(describeRunningEvent)
+              }
             )
           );
         }
@@ -4172,6 +4222,34 @@ function createToolUsageRule(options) {
       const names = tools.map(toolName);
       const nameSet = new Set(names);
       const findings = [];
+      const forbidden = new Set(options.forbidden ?? []);
+      const allowed = options.allowed?.length ? new Set(options.allowed) : void 0;
+      for (const event of tools) {
+        const name = toolName(event);
+        const display = formatToolNameForMessage(name);
+        if (forbidden.has(name)) {
+          findings.push(
+            failFinding(
+              "tool.usage",
+              `Forbidden tool ${display} appeared.`,
+              [eventEvidence3(event)],
+              "tool absent",
+              name
+            )
+          );
+        }
+        if (allowed && !allowed.has(name)) {
+          findings.push(
+            failFinding(
+              "tool.usage",
+              `Tool ${display} is not in the allowed tool set.`,
+              [eventEvidence3(event)],
+              [...allowed].sort(),
+              name
+            )
+          );
+        }
+      }
       for (const required of options.required ?? []) {
         if (!nameSet.has(required)) {
           const otherKinds = [
@@ -4189,7 +4267,7 @@ function createToolUsageRule(options) {
           findings.push(
             failFinding(
               "tool.usage",
-              `Required tool ${required} did not appear.${hint}`,
+              `Required tool ${formatToolNameForMessage(required)} did not appear.${hint}`,
               runEvidence(context.selectedRun),
               required,
               names
@@ -4197,21 +4275,6 @@ function createToolUsageRule(options) {
           );
         }
       }
-      const forbidden = new Set(options.forbidden ?? []);
-      const allowed = options.allowed?.length ? new Set(options.allowed) : void 0;
-      for (const event of tools) {
-        const name = toolName(event);
-        if (forbidden.has(name)) {
-          findings.push(
-            failFinding("tool.usage", `Forbidden tool ${name} appeared.`, [eventEvidence3(event)], "tool absent", name)
-          );
-        }
-        if (allowed && !allowed.has(name)) {
-          findings.push(
-            failFinding("tool.usage", `Tool ${name} is not in the allowed tool set.`, [eventEvidence3(event)], [...allowed].sort(), name)
-          );
-        }
-      }
       if (options.minCount !== void 0 && tools.length < options.minCount) {
         findings.push(
           failFinding("tool.usage", `Tool count ${tools.length} was below minimum ${options.minCount}.`, runEvidence(context.selectedRun), { minCount: options.minCount }, tools.length)
@@ -4245,7 +4308,36 @@ function createStepOrderingRule(options) {
       const afterEvents = finishedEvents(context, options.after.kind).filter(
         (event) => stepEndpointName(event, options.after.kind) === options.after.name
       );
-      if (beforeEvents.length === 0 || afterEvents.length === 0) {
+      const beforeRunning = semanticEvents(context).filter(
+        (event) => event.kind === options.before.kind && event.status === "running" && stepEndpointName(event, options.before.kind) === options.before.name
+      );
+      const afterRunning = semanticEvents(context).filter(
+        (event) => event.kind === options.after.kind && event.status === "running" && stepEndpointName(event, options.after.kind) === options.after.name
+      );
+      const afterStarted = afterEvents.length > 0 || afterRunning.length > 0;
+      const afterEvidenceEvent = afterEvents[0] ?? afterRunning[0];
+      if (beforeRunning.length > 0 && afterStarted && afterEvidenceEvent !== void 0) {
+        const endpointExpected2 = bothTools ? { before: options.before.name, after: options.after.name } : { before: options.before, after: options.after };
+        return [
+          failFinding(
+            ruleId,
+            `${subject} ${beforeLabel} started but never finished before ${afterLabel} ran.`,
+            [
+              ...beforeRunning.map((event) => eventEvidence3(event)),
+              eventEvidence3(afterEvidenceEvent)
+            ],
+            { ...endpointExpected2, mode },
+            {
+              code: bothTools ? "tool.order.interval-unresolved" : "step.order.interval-unresolved",
+              unfinishedBefore: true,
+              beforeStatus: "running",
+              ...beforeEvents.length > 0 ? { finishedBeforeMasked: true } : {},
+              ...afterEvents.length === 0 ? { afterStatus: "running" } : {}
+            }
+          )
+        ];
+      }
+      if (beforeEvents.length === 0 || !afterStarted) {
         if (!requireEndpoints) return [];
         const findings = [];
         if (beforeEvents.length === 0) {
@@ -4269,7 +4361,7 @@ function createStepOrderingRule(options) {
             )
           );
         }
-        if (afterEvents.length === 0) {
+        if (!afterStarted) {
           const otherKinds = otherKindsForStepName(
             context,
             options.after.kind,
@@ -4292,9 +4384,15 @@ function createStepOrderingRule(options) {
         }
         return findings;
       }
-      const encounter = finishedEvents(context).filter(
-        (event) => event.kind === options.before.kind && stepEndpointName(event, options.before.kind) === options.before.name || event.kind === options.after.kind && stepEndpointName(event, options.after.kind) === options.after.name
-      );
+      const afterForOrder = afterEvents.length > 0 ? afterEvents : afterRunning;
+      const encounter = [
+        ...finishedEvents(context).filter(
+          (event) => event.kind === options.before.kind && stepEndpointName(event, options.before.kind) === options.before.name || event.kind === options.after.kind && stepEndpointName(event, options.after.kind) === options.after.name
+        ),
+        ...afterEvents.length === 0 ? afterRunning.filter(
+          (event) => !finishedEvents(context).some((finished) => finished.eventId === event.eventId)
+        ) : []
+      ];
       let beforeIndex;
       let afterIndex;
       for (const [position, event] of encounter.entries()) {
@@ -4325,7 +4423,7 @@ function createStepOrderingRule(options) {
         ];
       }
       const beforeEvent = beforeEvents[0];
-      const afterEvent = afterEvents[0];
+      const afterEvent = afterForOrder[0];
       const beforeEnd = eventEndMs(beforeEvent);
       const afterStart = eventStartMs(afterEvent);
       const endpointExpected = bothTools ? { before: options.before.name, after: options.after.name } : { before: options.before, after: options.after };
@@ -4411,7 +4509,7 @@ function createStepOrderingRule(options) {
             latestBefore = { event, end };
           }
         }
-        for (const event of afterEvents) {
+        for (const event of afterForOrder) {
           const start = eventStartMs(event);
           if (start === void 0) {
             missingAfterStarts += 1;
@@ -4440,7 +4538,7 @@ function createStepOrderingRule(options) {
         const tie = tiedEnd === tiedStart ? recordedTieOrder(
           context,
           beforeEvents.filter((event) => eventEndMs(event) === tiedEnd),
-          afterEvents.filter((event) => eventStartMs(event) === tiedStart)
+          afterForOrder.filter((event) => eventStartMs(event) === tiedStart)
         ) : void 0;
         if (latestBefore.end > earliestAfter.start || tie !== void 0 && tie.inOrder !== true) {
           return [
@@ -4621,10 +4719,13 @@ function createStructureIncompleteRule(options = {}) {
           findings.push(
             failFinding(
               "structure.incomplete",
-              "Trace contains incomplete running events.",
+              incompleteRunningMessage("Trace contains incomplete running events", running),
               running.map((event) => eventEvidence3(event, "status")),
               "no running events",
-              running.length
+              {
+                count: running.length,
+                unfinished: running.map(describeRunningEvent)
+              }
             )
           );
         }
@@ -5418,5 +5519,5 @@ function runTraceChecks(input, options = {}) {
 }
 
 export { buildTraceFacts, createBaselineRegressionRule, createDecisionRule, createGuardrailRule, createLlmUsageRule, createMaxStepDurationRule, createObservedOutcomeRule, createRequireCompletedRule, createRetrievalRule, createRunDepthRule, createRunDurationRule, createRunEventCountRule, createRunStatusRule, createSafetyOversizedAttributeRule, createSafetyRawContentRule, createSafetyRedactionRule, createSafetySecretPatternRule, createStallDetectionRule, createStepOrderingRule, createStructureCycleRule, createStructureIncompleteRule, createStructureOrphanRule, createStructureParallelWidthRule, createStructureRelationshipRule, createToolFailureRule, createToolOrderingRule, createToolUsageRule, defineTraceContract, deriveFailureFacts, deriveRelationshipFacts, evaluateToolArgumentValue, evaluateTraceContract, evaluateTraceContractRead, explainTraceContract, extractSessionWorkflowMetadata, extractToolArgumentPayload, lintTraceContract, resolveJsonPointer, resolveTraceContractScope, runTraceChecks, sessionKeyForRun, summarizeSemanticParity, workflowMetadataForRun };
-//# sourceMappingURL=chunk-6G67ZN72.mjs.map
-//# sourceMappingURL=chunk-6G67ZN72.mjs.map
+//# sourceMappingURL=chunk-CXC7L4PY.mjs.map
+//# sourceMappingURL=chunk-CXC7L4PY.mjs.map