@agentv/core 3.12.0 → 3.13.1

package/dist/evaluation/validation/index.js

@@ -8,7 +8,7 @@ import {
   isEvaluatorKind,
   loadCasesFromFile,
   resolveFileReference
-} from "../../chunk-4XWPXNQM.js";
+} from "../../chunk-ZB3AUPES.js";
 
 // src/evaluation/validation/file-type.ts
 import { readFile } from "node:fs/promises";

package/dist/index.cjs

@@ -1380,132 +1380,6 @@ var init_otlp_json_file_exporter = __esm({
   }
 });
 
-// src/observability/simple-trace-file-exporter.ts
-var simple_trace_file_exporter_exports = {};
-__export(simple_trace_file_exporter_exports, {
-  SimpleTraceFileExporter: () => SimpleTraceFileExporter
-});
-function hrTimeDiffMs(start, end) {
-  const diffSec = end[0] - start[0];
-  const diffNano = end[1] - start[1];
-  return Math.round(diffSec * 1e3 + diffNano / 1e6);
-}
-var import_node_fs16, import_promises34, import_node_path50, SimpleTraceFileExporter;
-var init_simple_trace_file_exporter = __esm({
-  "src/observability/simple-trace-file-exporter.ts"() {
-    "use strict";
-    import_node_fs16 = require("fs");
-    import_promises34 = require("fs/promises");
-    import_node_path50 = require("path");
-    SimpleTraceFileExporter = class {
-      stream = null;
-      filePath;
-      streamReady = null;
-      pendingWrites = [];
-      _shuttingDown = false;
-      spansByTraceId = /* @__PURE__ */ new Map();
-      constructor(filePath) {
-        this.filePath = filePath;
-      }
-      async ensureStream() {
-        if (!this.streamReady) {
-          this.streamReady = (async () => {
-            await (0, import_promises34.mkdir)((0, import_node_path50.dirname)(this.filePath), { recursive: true });
-            this.stream = (0, import_node_fs16.createWriteStream)(this.filePath, { flags: "w" });
-            return this.stream;
-          })();
-        }
-        return this.streamReady;
-      }
-      export(spans, resultCallback) {
-        if (this._shuttingDown) {
-          resultCallback({ code: 0 });
-          return;
-        }
-        const rootSpans = [];
-        for (const span of spans) {
-          const traceId = span.spanContext().traceId;
-          const existing = this.spansByTraceId.get(traceId) ?? [];
-          existing.push(span);
-          this.spansByTraceId.set(traceId, existing);
-          if (span.name === "agentv.eval") {
-            rootSpans.push(span);
-          }
-        }
-        const writePromise = this.ensureStream().then((stream) => {
-          for (const root of rootSpans) {
-            const traceId = root.spanContext().traceId;
-            const traceSpans = this.spansByTraceId.get(traceId) ?? [root];
-            const children = traceSpans.filter(
-              (span) => span.spanContext().spanId !== root.spanContext().spanId
-            );
-            const record = this.buildSimpleRecord(root, children);
-            stream.write(`${JSON.stringify(record)}
-`);
-            this.spansByTraceId.delete(traceId);
-          }
-        });
-        this.pendingWrites.push(writePromise);
-        resultCallback({ code: 0 });
-      }
-      async shutdown() {
-        this._shuttingDown = true;
-        await Promise.all(this.pendingWrites);
-        this.pendingWrites = [];
-        this.spansByTraceId.clear();
-        return new Promise((resolve) => {
-          if (this.stream) {
-            this.stream.end(() => resolve());
-          } else {
-            resolve();
-          }
-        });
-      }
-      async forceFlush() {
-        await Promise.all(this.pendingWrites);
-        this.pendingWrites = [];
-      }
-      buildSimpleRecord(root, children) {
-        const attrs = root.attributes || {};
-        const durationMs = typeof attrs["agentv.trace.duration_ms"] === "number" ? attrs["agentv.trace.duration_ms"] : hrTimeDiffMs(root.startTime, root.endTime);
-        let inputTokens = 0;
-        let outputTokens = 0;
-        for (const child of children) {
-          const ca = child.attributes || {};
-          if (ca["gen_ai.usage.input_tokens"]) inputTokens += ca["gen_ai.usage.input_tokens"];
-          if (ca["gen_ai.usage.output_tokens"]) outputTokens += ca["gen_ai.usage.output_tokens"];
-        }
-        const rootInputTokens = typeof attrs["agentv.trace.token_input"] === "number" ? attrs["agentv.trace.token_input"] : 0;
-        const rootOutputTokens = typeof attrs["agentv.trace.token_output"] === "number" ? attrs["agentv.trace.token_output"] : 0;
-        const rootCachedTokens = typeof attrs["agentv.trace.token_cached"] === "number" ? attrs["agentv.trace.token_cached"] : void 0;
-        const llmSpans = children.filter((s) => s.attributes?.["gen_ai.operation.name"] === "chat").map((s) => ({
-          type: "llm",
-          name: s.name,
-          duration_ms: hrTimeDiffMs(s.startTime, s.endTime)
-        }));
-        const toolSpans = children.filter((s) => s.attributes?.["gen_ai.tool.name"]).map((s) => ({
-          type: "tool",
-          name: s.attributes["gen_ai.tool.name"],
-          duration_ms: hrTimeDiffMs(s.startTime, s.endTime)
-        }));
-        return {
-          test_id: attrs["agentv.test_id"],
-          target: attrs["agentv.target"],
-          score: attrs["agentv.score"],
-          duration_ms: durationMs,
-          cost_usd: attrs["agentv.trace.cost_usd"],
-          token_usage: inputTokens || outputTokens || rootInputTokens || rootOutputTokens || rootCachedTokens ? {
-            input: inputTokens || rootInputTokens,
-            output: outputTokens || rootOutputTokens,
-            ...rootCachedTokens ? { cached: rootCachedTokens } : {}
-          } : void 0,
-          spans: [...llmSpans, ...toolSpans].length > 0 ? [...llmSpans, ...toolSpans] : void 0
-        };
-      }
-    };
-  }
-});
-
 // src/index.ts
 var index_exports = {};
 __export(index_exports, {
@@ -1529,7 +1403,6 @@ __export(index_exports, {
   ProviderRegistry: () => ProviderRegistry,
   RepoManager: () => RepoManager,
   ResponseCache: () => ResponseCache,
-  SimpleTraceFileExporter: () => SimpleTraceFileExporter,
   SkillTriggerEvaluator: () => SkillTriggerEvaluator,
   TEST_MESSAGE_ROLES: () => TEST_MESSAGE_ROLES,
   TemplateNotDirectoryError: () => TemplateNotDirectoryError,
@@ -1708,8 +1581,6 @@ function isTestMessage(value) {
 var EVALUATOR_KIND_VALUES = [
   "code-grader",
   "llm-grader",
-  "code-judge",
-  "llm-judge",
   "rubric",
   "composite",
   "tool-trajectory",
@@ -2460,12 +2331,6 @@ function parseExecutionDefaults(raw, configPath) {
   } else if (obj.verbose !== void 0) {
     logWarning(`Invalid execution.verbose in ${configPath}, expected boolean`);
   }
-  const traceFile = obj.trace_file;
-  if (typeof traceFile === "string" && traceFile.trim().length > 0) {
-    result.trace_file = traceFile.trim();
-  } else if (traceFile !== void 0) {
-    logWarning(`Invalid execution.trace_file in ${configPath}, expected non-empty string`);
-  }
   if (typeof obj.keep_workspaces === "boolean") {
     result.keep_workspaces = obj.keep_workspaces;
   } else if (obj.keep_workspaces !== void 0) {
@@ -2582,6 +2447,9 @@ var ANSI_RESET5 = "\x1B[0m";
 function normalizeEvaluatorType(type) {
   return type.replace(/_/g, "-");
 }
+function isDeprecatedJudgeType(type) {
+  return type === "code-judge" || type === "llm-judge";
+}
 async function parseEvaluators(rawEvalCase, globalExecution, searchRoots, evalId) {
   const execution = rawEvalCase.execution;
   const executionObject = isJsonObject2(execution) ? execution : void 0;
@@ -2644,6 +2512,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
     const rawName = asString(rawEvaluator.name);
     const rawType = rawEvaluator.type;
     const typeValue = typeof rawType === "string" ? normalizeEvaluatorType(rawType) : rawType;
+    if (typeof typeValue === "string" && isDeprecatedJudgeType(typeValue)) {
+      logWarning2(
+        `Skipping evaluator '${rawName ?? "<unnamed>"}' in '${evalId}': '${rawType}' is deprecated. Use '${typeValue.replace("-judge", "-grader")}' instead`
+      );
+      continue;
+    }
     const isCustomType = typeof typeValue === "string" && !isEvaluatorKind(typeValue);
     if (typeof typeValue !== "string") {
       logWarning2(`Skipping evaluator with invalid type in '${evalId}'`);
@@ -2676,7 +2550,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       });
       continue;
     }
-    if (typeValue === "code-grader" || typeValue === "code-judge") {
+    if (typeValue === "code-grader") {
       let command;
       if (rawEvaluator.script !== void 0 && rawEvaluator.command === void 0) {
         console.warn(
@@ -2786,7 +2660,14 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const aggregatorType = asString(rawAggregator.type);
-      if (aggregatorType !== "weighted_average" && aggregatorType !== "code-grader" && aggregatorType !== "code-judge" && aggregatorType !== "llm-grader" && aggregatorType !== "llm-judge" && aggregatorType !== "threshold") {
+      const normalizedAggregatorType = typeof aggregatorType === "string" ? aggregatorType === "weighted_average" || aggregatorType === "threshold" ? aggregatorType : normalizeEvaluatorType(aggregatorType) : aggregatorType;
+      if (typeof normalizedAggregatorType === "string" && isDeprecatedJudgeType(normalizedAggregatorType)) {
+        logWarning2(
+          `Skipping composite evaluator '${name}' in '${evalId}': aggregator type '${aggregatorType}' is deprecated. Use '${normalizedAggregatorType.replace("-judge", "-grader")}' instead`
+        );
+        continue;
+      }
+      if (normalizedAggregatorType !== "weighted_average" && normalizedAggregatorType !== "code-grader" && normalizedAggregatorType !== "llm-grader" && normalizedAggregatorType !== "threshold") {
         logWarning2(
           `Skipping composite evaluator '${name}' in '${evalId}': invalid aggregator type '${aggregatorType}'`
         );
@@ -2821,7 +2702,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       let aggregator;
-      if (aggregatorType === "weighted_average") {
+      if (normalizedAggregatorType === "weighted_average") {
         const weights = isJsonObject2(rawAggregator.weights) ? rawAggregator.weights : void 0;
         const parsedWeights = {};
         if (weights) {
@@ -2835,7 +2716,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
           type: "weighted_average",
           ...Object.keys(parsedWeights).length > 0 ? { weights: parsedWeights } : {}
         };
-      } else if (aggregatorType === "code-grader" || aggregatorType === "code-judge") {
+      } else if (normalizedAggregatorType === "code-grader") {
         const aggregatorPath = asString(rawAggregator.path);
         if (!aggregatorPath) {
           logWarning2(
@@ -2848,7 +2729,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
           path: aggregatorPath,
           cwd: searchRoots[0]
         };
-      } else if (aggregatorType === "threshold") {
+      } else if (normalizedAggregatorType === "threshold") {
         const thresholdValue = rawAggregator.threshold;
         if (typeof thresholdValue !== "number" || thresholdValue < 0 || thresholdValue > 1) {
           logWarning2(
@@ -3596,10 +3477,15 @@ function coerceEvaluator(candidate, contextId) {
     return void 0;
   }
   const normalized = normalizeEvaluatorType(candidate);
+  if (isDeprecatedJudgeType(normalized)) {
+    throw new Error(
+      `Unsupported grader '${candidate}' in ${contextId}. Use '${normalized.replace("-judge", "-grader")}' instead.`
+    );
+  }
   if (isEvaluatorKind(normalized)) {
     return normalized;
   }
-  logWarning2(`Unknown evaluator '${candidate}' in ${contextId}, falling back to default`);
+  logWarning2(`Unknown grader '${candidate}' in ${contextId}, falling back to default`);
   return void 0;
 }
 function asString(value) {
@@ -5032,9 +4918,7 @@ function assertionToNaturalLanguage(entry) {
     case "ends_with":
       return `Output ends with '${entry.value}'`;
     case "llm-grader":
-    case "llm_grader":
-    case "llm-judge":
-    case "llm_judge": {
+    case "llm_grader": {
       if (Array.isArray(entry.rubrics) && entry.rubrics.length > 0) {
         return null;
       }
@@ -5047,9 +4931,7 @@ function assertionToNaturalLanguage(entry) {
       return tools ? `Agent called tools in order: ${tools}` : "Agent followed expected tool trajectory";
     }
     case "code-grader":
-    case "code_grader":
-    case "code-judge":
-    case "code_judge": {
+    case "code_grader": {
       const graderName = entry.name ?? deriveGraderNameFromCommand(entry.command) ?? "code-grader";
       const desc = typeof entry.description === "string" ? entry.description : void 0;
       return codeGraderInstruction(graderName, desc);
@@ -5080,7 +4962,7 @@ function assertionToNaturalLanguage(entry) {
   }
 }
 function assertionToNaturalLanguageList(entry) {
-  if (entry.type === "llm-grader" || entry.type === "llm_grader" || entry.type === "llm-judge" || entry.type === "llm_judge") {
+  if (entry.type === "llm-grader" || entry.type === "llm_grader") {
     if (Array.isArray(entry.rubrics) && entry.rubrics.length > 0) {
       return entry.rubrics.map((r) => r.outcome ?? r.criteria ?? r.id).filter((s) => typeof s === "string");
     }
@@ -12810,10 +12692,26 @@ function extractJsonBlob(text) {
   const match = text.match(/\{[\s\S]*\}/);
   return match?.[0];
 }
+function repairSchemaNearBooleanFields(text) {
+  return text.replace(
+    /("passed"\s*:\s*)(?:"([^"]+)"|([A-Za-z_][A-Za-z0-9_-]*))/gi,
+    (_match, prefix, quotedValue, bareValue) => {
+      const value = (quotedValue ?? bareValue ?? "").trim().toLowerCase();
+      if (value === "true") {
+        return `${prefix}true`;
+      }
+      if (value === "false") {
+        return `${prefix}false`;
+      }
+      return `${prefix}false`;
+    }
+  );
+}
 function parseJsonFromText(text) {
   const cleaned = typeof text === "string" ? text.replace(/```json\n?|```/g, "").trim() : "";
   const blob = extractJsonBlob(cleaned) ?? cleaned;
-  return JSON.parse(blob);
+  const repaired = repairSchemaNearBooleanFields(blob);
+  return JSON.parse(repaired);
 }
 function isNonEmptyString(value) {
   return typeof value === "string" && value.trim().length > 0;
@@ -12960,12 +12858,12 @@ async function execFileWithStdinNode(argv, stdinPayload, options) {
   });
 }
 async function execShellWithStdin(command, stdinPayload, options = {}) {
-  const { mkdir: mkdir18, readFile: readFile14, rm: rm6, writeFile: writeFile9 } = await import("fs/promises");
+  const { mkdir: mkdir17, readFile: readFile14, rm: rm6, writeFile: writeFile9 } = await import("fs/promises");
   const { tmpdir: tmpdir3 } = await import("os");
   const path48 = await import("path");
   const { randomUUID: randomUUID10 } = await import("crypto");
   const dir = path48.join(tmpdir3(), `agentv-exec-${randomUUID10()}`);
-  await mkdir18(dir, { recursive: true });
+  await mkdir17(dir, { recursive: true });
   const stdinPath = path48.join(dir, "stdin.txt");
   const stdoutPath = path48.join(dir, "stdout.txt");
   const stderrPath = path48.join(dir, "stderr.txt");
@@ -13285,7 +13183,7 @@ function toCamelCaseDeep(obj) {
 // src/evaluation/evaluators/code-evaluator.ts
 var FILE_BACKED_OUTPUT_THRESHOLD = 5e4;
 var CodeEvaluator = class {
-  kind = "code-judge";
+  kind = "code-grader";
   command;
   cwd;
   agentTimeoutMs;
@@ -13304,7 +13202,7 @@ var CodeEvaluator = class {
     if (outputForPayload) {
       const serialized = JSON.stringify(outputForPayload);
       if (serialized.length > FILE_BACKED_OUTPUT_THRESHOLD) {
-        const tmpDir = await (0, import_promises26.mkdtemp)((0, import_node_path36.join)((0, import_node_os5.tmpdir)(), "agentv-judge-"));
+        const tmpDir = await (0, import_promises26.mkdtemp)((0, import_node_path36.join)((0, import_node_os5.tmpdir)(), "agentv-grader-"));
         outputPath = (0, import_node_path36.join)(tmpDir, "output.json");
         await (0, import_promises26.writeFile)(outputPath, serialized);
         outputForPayload = null;
@@ -13594,7 +13492,7 @@ var LlmGraderEvaluator = class {
       return this.evaluateWithDelegatedAgent(context2, graderProvider);
     }
     const config = context2.evaluator;
-    if ((config?.type === "llm-grader" || config?.type === "llm-judge") && config.rubrics && config.rubrics.length > 0) {
+    if (config?.type === "llm-grader" && config.rubrics && config.rubrics.length > 0) {
       return this.evaluateWithRubrics(context2, graderProvider, config.rubrics);
     }
     return this.evaluateFreeform(context2, graderProvider);
@@ -13779,7 +13677,7 @@ ${context2.fileChanges}`;
     const systemPrompt = this.buildAgentSystemPrompt(context2);
     const userPrompt = this.buildAgentUserPrompt(context2);
     const config = context2.evaluator;
-    const rubrics = config?.type === "llm-grader" || config?.type === "llm-judge" ? config.rubrics : void 0;
+    const rubrics = config?.type === "llm-grader" ? config.rubrics : void 0;
     const fsTools = createFilesystemTools(workspacePath);
     const evaluatorRawRequest = {
       mode: "built-in",
@@ -13875,7 +13773,7 @@ ${context2.fileChanges}`;
         };
       }
       const config = context2.evaluator;
-      const rubrics = config?.type === "llm-grader" || config?.type === "llm-judge" ? config.rubrics : void 0;
+      const rubrics = config?.type === "llm-grader" ? config.rubrics : void 0;
       const details = {
         mode: modeLabel,
         grader_target: provider.targetName
@@ -13915,7 +13813,7 @@ ${context2.fileChanges}`;
    */
   buildAgentSystemPrompt(context2) {
     const config = context2.evaluator;
-    const rubrics = config?.type === "llm-grader" || config?.type === "llm-judge" ? config.rubrics : void 0;
+    const rubrics = config?.type === "llm-grader" ? config.rubrics : void 0;
     const parts = [
       "You are an expert evaluator with access to the workspace filesystem.",
       "Use the provided tools to investigate the workspace and verify the criteria are met.",
@@ -13946,7 +13844,7 @@ ${context2.fileChanges}`;
       return substituteVariables(this.evaluatorTemplate, variables);
     }
     const config = context2.evaluator;
-    const rubrics = config?.type === "llm-grader" || config?.type === "llm-judge" ? config.rubrics : void 0;
+    const rubrics = config?.type === "llm-grader" ? config.rubrics : void 0;
     const parts = [
       "Evaluate the candidate answer by investigating the workspace.",
       "",
@@ -13989,7 +13887,7 @@ ${context2.fileChanges}`;
   buildDelegatedPrompt(context2) {
     const formattedQuestion = context2.promptInputs.question && context2.promptInputs.question.trim().length > 0 ? context2.promptInputs.question : context2.evalCase.question;
     const config = context2.evaluator;
-    const rubrics = config?.type === "llm-grader" || config?.type === "llm-judge" ? config.rubrics : void 0;
+    const rubrics = config?.type === "llm-grader" ? config.rubrics : void 0;
     if (this.evaluatorTemplate) {
       const variables = {
         [TEMPLATE_VARIABLES.CRITERIA]: context2.evalCase.criteria.trim(),
@@ -14486,10 +14384,8 @@ var CompositeEvaluator = class {
     const aggregator = this.config.aggregator;
     switch (aggregator.type) {
       case "code-grader":
-      case "code-judge":
         return this.runCodeAggregator(results, aggregator.path, aggregator.cwd ?? this.cwd);
       case "llm-grader":
-      case "llm-judge":
         return this.runLlmAggregator(results, context2, aggregator);
       case "threshold":
         return this.runThreshold(results, aggregator.threshold);
@@ -16911,7 +16807,7 @@ var endsWithFactory = (config) => {
 };
 function createBuiltinRegistry() {
   const registry = new EvaluatorRegistry();
-  registry.register("llm-grader", llmGraderFactory).register("llm-judge", llmGraderFactory).register("code-grader", codeFactory).register("code-judge", codeFactory).register("composite", compositeFactory).register("tool-trajectory", toolTrajectoryFactory).register("field-accuracy", fieldAccuracyFactory).register("latency", latencyFactory).register("cost", costFactory).register("token-usage", tokenUsageFactory).register("execution-metrics", executionMetricsFactory).register("skill-trigger", skillTriggerFactory).register("contains", containsFactory).register("contains-any", containsAnyFactory).register("contains-all", containsAllFactory).register("icontains", icontainsFactory).register("icontains-any", icontainsAnyFactory).register("icontains-all", icontainsAllFactory).register("starts-with", startsWithFactory).register("ends-with", endsWithFactory).register("regex", regexFactory).register("is-json", isJsonFactory).register("equals", equalsFactory).register("inline-assert", (config) => {
+  registry.register("llm-grader", llmGraderFactory).register("code-grader", codeFactory).register("composite", compositeFactory).register("tool-trajectory", toolTrajectoryFactory).register("field-accuracy", fieldAccuracyFactory).register("latency", latencyFactory).register("cost", costFactory).register("token-usage", tokenUsageFactory).register("execution-metrics", executionMetricsFactory).register("skill-trigger", skillTriggerFactory).register("contains", containsFactory).register("contains-any", containsAnyFactory).register("contains-all", containsAllFactory).register("icontains", icontainsFactory).register("icontains-any", icontainsAnyFactory).register("icontains-all", icontainsAllFactory).register("starts-with", startsWithFactory).register("ends-with", endsWithFactory).register("regex", regexFactory).register("is-json", isJsonFactory).register("equals", equalsFactory).register("inline-assert", (config) => {
     const fn = config[INLINE_ASSERT_FN];
     if (!fn) {
       throw new Error(
@@ -19629,7 +19525,7 @@ function filterEvalCases(evalCases, filter) {
   return evalCases.filter((evalCase) => import_micromatch3.default.isMatch(evalCase.id, filter));
 }
 function buildEvaluatorRegistry(overrides, resolveGraderProvider) {
-  const llmGrader = overrides?.["llm-grader"] ?? overrides?.["llm-judge"] ?? new LlmGraderEvaluator({
+  const llmGrader = overrides?.["llm-grader"] ?? new LlmGraderEvaluator({
     resolveGraderProvider: async (context2) => {
       if (context2.graderProvider) {
         return context2.graderProvider;
@@ -20061,8 +19957,6 @@ var AgentVConfigSchema = import_zod5.z.object({
     agentTimeoutMs: import_zod5.z.number().int().min(0).optional(),
     /** Enable verbose logging */
     verbose: import_zod5.z.boolean().optional(),
-    /** Write human-readable trace JSONL to this path (supports {timestamp} placeholder) */
-    traceFile: import_zod5.z.string().optional(),
     /** Always keep temp workspaces after eval */
     keepWorkspaces: import_zod5.z.boolean().optional(),
     /** Write OTLP JSON trace to this path (supports {timestamp} placeholder) */
@@ -20362,12 +20256,6 @@ var OtelTraceExporter = class {
           new SimpleSpanProcessor(new OtlpJsonFileExporter2(this.options.otlpFilePath))
         );
       }
-      if (this.options.traceFilePath) {
-        const { SimpleTraceFileExporter: SimpleTraceFileExporter2 } = await Promise.resolve().then(() => (init_simple_trace_file_exporter(), simple_trace_file_exporter_exports));
-        processors.push(
-          new SimpleSpanProcessor(new SimpleTraceFileExporter2(this.options.traceFilePath))
-        );
-      }
       if (processors.length === 0) {
         return false;
       }
@@ -20481,10 +20369,10 @@ var OtelTraceExporter = class {
         }
         if (result.scores) {
           for (const score of result.scores) {
-            rootSpan.addEvent(`agentv.evaluator.${score.name}`, {
-              "agentv.evaluator.score": score.score,
-              "agentv.evaluator.type": score.type,
-              ...score.verdict ? { "agentv.evaluator.verdict": score.verdict } : {}
+            rootSpan.addEvent(`agentv.grader.${score.name}`, {
+              "agentv.grader.score": score.score,
+              "agentv.grader.type": score.type,
+              ...score.verdict ? { "agentv.grader.verdict": score.verdict } : {}
             });
           }
         }
@@ -20795,7 +20683,6 @@ function toHrTime(iso) {
 
 // src/observability/index.ts
 init_otlp_json_file_exporter();
-init_simple_trace_file_exporter();
 
 // src/index.ts
 function createAgentKernel() {
@@ -20823,7 +20710,6 @@ function createAgentKernel() {
   ProviderRegistry,
   RepoManager,
   ResponseCache,
-  SimpleTraceFileExporter,
   SkillTriggerEvaluator,
   TEST_MESSAGE_ROLES,
   TemplateNotDirectoryError,