agent-inspect 6.7.2 → 6.7.3

package/packages/cli/dist/index.cjs

@@ -8489,7 +8489,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average duration (ms)",
           baselineAgg?.avgDurationMs,
-          candidateAgg?.avgDurationMs
+          candidateAgg?.avgDurationMs,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8499,7 +8501,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average LLM calls",
           baselineAgg?.avgLlmCallCount,
-          candidateAgg?.avgLlmCallCount
+          candidateAgg?.avgLlmCallCount,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8509,7 +8513,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average token usage",
           baselineAgg?.avgTokenUsage,
-          candidateAgg?.avgTokenUsage
+          candidateAgg?.avgTokenUsage,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8519,7 +8525,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Average retries",
           baselineAgg?.avgRetryCount,
-          candidateAgg?.avgRetryCount
+          candidateAgg?.avgRetryCount,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8529,7 +8537,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Observation failure rate",
           baselineAgg?.observationFailureRate,
-          candidateAgg?.observationFailureRate
+          candidateAgg?.observationFailureRate,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8565,7 +8575,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Guardrail failures",
           baselineAgg?.avgGuardrailFailures,
-          candidateAgg?.avgGuardrailFailures
+          candidateAgg?.avgGuardrailFailures,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8575,7 +8587,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Circuit violations",
           baselineAgg?.avgCircuitViolations,
-          candidateAgg?.avgCircuitViolations
+          candidateAgg?.avgCircuitViolations,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -8585,7 +8599,9 @@ function compareCohortAggregates(groups, options) {
           metric,
           "Redaction warnings",
           baselineAgg?.avgRedactionWarnings,
-          candidateAgg?.avgRedactionWarnings
+          candidateAgg?.avgRedactionWarnings,
+          true,
+          tolerance
         );
         if (item) comparisons.push(item);
         break;
@@ -9677,7 +9693,7 @@ var require_bindings = __commonJS({
   "node_modules/.pnpm/bindings@1.5.0/node_modules/bindings/bindings.js"(exports$1, module) {
     var fs = __require("fs");
     var path37 = __require("path");
-    var fileURLToPath2 = require_file_uri_to_path();
+    var fileURLToPath3 = require_file_uri_to_path();
     var join = path37.join;
     var dirname = path37.dirname;
     var exists2 = fs.accessSync && function(path38) {
@@ -9789,7 +9805,7 @@ var require_bindings = __commonJS({
       Error.stackTraceLimit = origSTL;
       var fileSchema = "file://";
       if (fileName.indexOf(fileSchema) === 0) {
-        fileName = fileURLToPath2(fileName);
+        fileName = fileURLToPath3(fileName);
       }
       return fileName;
     };
@@ -10788,7 +10804,7 @@ var init_src = __esm({
 });
 
 // package.json
-var version = "6.7.2";
+var version = "6.7.3";
 
 // packages/cli/src/list.ts
 init_advanced();
@@ -11749,13 +11765,14 @@ function parseTimestamp(v) {
 function hasToken(hay, token) {
   return hay.includes(token);
 }
+var ERROR_TOKEN_RE = /\.(error|failed)(\.|$)/;
 function inferKind(eventName) {
   if (hasToken(eventName, ".llm.")) return "LLM";
   if (hasToken(eventName, ".tool.")) return "TOOL";
   if (hasToken(eventName, ".agent.")) return "AGENT";
   if (hasToken(eventName, ".retriever.")) return "RETRIEVER";
   if (hasToken(eventName, ".result.")) return "RESULT";
-  if (eventName.endsWith(".error") || eventName.endsWith(".failed") || eventName.includes(".error") || eventName.includes(".failed")) {
+  if (ERROR_TOKEN_RE.test(eventName)) {
     return "ERROR";
   }
   return "LOG";
@@ -11835,7 +11852,7 @@ var EventNormalizer = class {
       status = mapping.status;
     } else if (mapping?.kind === "ERROR") {
       status = "error";
-    } else if (eventName.includes(".failed") || eventName.includes(".error")) {
+    } else if (ERROR_TOKEN_RE.test(eventName)) {
       status = "error";
     } else if (mapping?.startsRun || mapping?.startsStep) {
       status = "running";
@@ -13038,6 +13055,11 @@ init_helpers();
 function hexFrom(seed, byteLen) {
   return crypto__default.default.createHash("sha256").update(seed, "utf8").digest("hex").slice(0, byteLen * 2);
 }
+function unixNano(ms) {
+  const whole = Math.trunc(ms);
+  const fractionNs = Math.round((ms - whole) * 1e6);
+  return BigInt(whole) * 1000000n + BigInt(fractionNs);
+}
 function mapInspectKindToOI(kind, warnings) {
   switch (kind) {
     case "LLM":
@@ -13087,10 +13109,10 @@ function exportOpenInference(tree, options) {
     const ev = n.event;
     const spanId = hexFrom(`${tree.runId}:${ev.eventId}`, 8);
     const parentSpanHex = ev.parentId ? hexFrom(`${tree.runId}:${ev.parentId}`, 8) : void 0;
-    const startNs = Math.round(ev.timestamp * 1e6);
+    const startNs = unixNano(ev.timestamp);
     let endNs;
     if (ev.durationMs !== void 0 && Number.isFinite(ev.durationMs)) {
-      endNs = startNs + Math.round(ev.durationMs * 1e6);
+      endNs = startNs + unixNano(ev.durationMs);
     }
     const { openInferenceKind } = mapInspectKindToOI(ev.kind, warnings);
     const attrs = {
@@ -13138,8 +13160,8 @@ function exportOpenInference(tree, options) {
       span_id: spanId,
       parent_span_id: parentSpanHex,
       name: ev.name,
-      start_time_unix_nano: startNs,
-      end_time_unix_nano: endNs,
+      start_time_unix_nano: startNs.toString(),
+      end_time_unix_nano: endNs?.toString(),
       attributes: attrs,
       status
     });
@@ -13891,6 +13913,10 @@ function manualTraceEventsToComparableRun(events) {
       meta2 = { ...meta2 ?? {}, agent_inspect_diff_parent_missing: true };
     }
     const outputPreview = extractOutputPreview(meta2);
+    if (meta2 !== void 0 && ("outputPreview" in meta2 || "resultPreview" in meta2)) {
+      delete meta2.outputPreview;
+      delete meta2.resultPreview;
+    }
     const sc = {
       id: acc.id,
       name: acc.name,
@@ -14540,6 +14566,7 @@ async function searchCommand(options = {}) {
           console.log(`Session not found: ${sessionId}`);
           console.log(`Trace directory: ${traceDir}`);
         }
+        process.exitCode = 1;
         return;
       }
       metas = scoped.metas;
@@ -18127,13 +18154,11 @@ async function studioCommand(options = {}) {
   await new Promise(() => {
   });
 }
-
-// packages/eval/src/index.ts
 function isTraceReadResult(value) {
   return value !== null && typeof value === "object" && "runs" in value && "events" in value && "format" in value;
 }
 function traceInputFrom(value) {
-  return { type: "file", path: value instanceof URL ? value.pathname : value };
+  return { type: "file", path: value instanceof URL ? url.fileURLToPath(value) : value };
 }
 async function resolveRead(input3, options) {
   try {
@@ -20457,7 +20482,7 @@ jobs:
 async function planInit(options = {}) {
   const framework = normalizeFramework(options.framework);
   const cwd = path26__default.default.resolve(options.cwd ?? process.cwd());
-  const demoPath = framework === "custom" ? path26__default.default.join("examples", "agent-inspect-demo.mjs") : path26__default.default.join("examples", `agent-inspect-${framework}-demo.mjs`);
+  const demoPath = framework === "custom" ? "examples/agent-inspect-demo.mjs" : `examples/agent-inspect-${framework}-demo.mjs`;
   const candidates = [
     { rel: CONFIG_FILE, content: configTemplate(framework) },
     { rel: GITKEEP, content: "" },