@agent-inspect/viewer 6.31.31 → 6.31.33

package/dist/index.cjs

@@ -2226,6 +2226,11 @@ 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));
@@ -2344,6 +2349,43 @@ 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 ${stepEndpointName(event)}`;
+  }
+  return `${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) {
+  for (const prefix of ["llm:", "generation:", "transcription:", "speech:"]) {
+    if (event.name.startsWith(prefix)) return event.name.slice(prefix.length);
+  }
+  return event.name;
+}
 function semanticEvents(context) {
   return context.logicalEvents ?? context.events;
 }
@@ -2397,10 +2439,13 @@ function createRunStatusRule(options = {}) {
           findings.push(
             failFinding(
               "run.status",
-              "Run contains incomplete running events.",
+              incompleteRunningMessage("Run contains incomplete running events", running),
               running.map((event) => eventEvidence(event)),
               "no running events",
-              running.length
+              {
+                count: running.length,
+                unfinished: running.map(describeRunningEvent)
+              }
             )
           );
         }
@@ -2468,6 +2513,34 @@ 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.`,
+              [eventEvidence(event)],
+              "tool absent",
+              name
+            )
+          );
+        }
+        if (allowed && !allowed.has(name)) {
+          findings.push(
+            failFinding(
+              "tool.usage",
+              `Tool ${display} is not in the allowed tool set.`,
+              [eventEvidence(event)],
+              [...allowed].sort(),
+              name
+            )
+          );
+        }
+      }
       for (const required of options.required ?? []) {
         if (!nameSet.has(required)) {
           const otherKinds = [
@@ -2485,7 +2558,7 @@ 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
@@ -2493,21 +2566,6 @@ 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.`, [eventEvidence(event)], "tool absent", name)
-          );
-        }
-        if (allowed && !allowed.has(name)) {
-          findings.push(
-            failFinding("tool.usage", `Tool ${name} is not in the allowed tool set.`, [eventEvidence(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)