agentv 3.5.0 → 3.7.0

package/dist/{chunk-5GG6DDP5.js → chunk-TR6H437M.js}

@@ -16,7 +16,7 @@ import {
   validateEvalFile,
   validateFileReferences,
   validateTargetsFile
-} from "./chunk-RLL4QGNL.js";
+} from "./chunk-7YS6YNJZ.js";
 import {
   createBuiltinRegistry,
   createProvider,
@@ -34,7 +34,7 @@ import {
   toSnakeCaseDeep as toSnakeCaseDeep2,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-D6G4N2H2.js";
+} from "./chunk-XGG64VIY.js";
 import {
   __commonJS,
   __esm,
@@ -3493,9 +3493,9 @@ var ASSERTION_TEMPLATES = {
   default: `#!/usr/bin/env bun
 import { defineAssertion } from '@agentv/eval';
 
-export default defineAssertion(({ answer }) => {
+export default defineAssertion(({ outputText }) => {
   // TODO: Implement your assertion logic
-  const pass = answer.length > 0;
+  const pass = outputText.length > 0;
   return {
     pass,
     reasoning: pass ? 'Output has content' : 'Output is empty',
@@ -3505,9 +3505,9 @@ export default defineAssertion(({ answer }) => {
   score: `#!/usr/bin/env bun
 import { defineAssertion } from '@agentv/eval';
 
-export default defineAssertion(({ answer }) => {
+export default defineAssertion(({ outputText }) => {
   // TODO: Implement your scoring logic (0.0 to 1.0)
-  const score = answer.length > 0 ? 1.0 : 0.0;
+  const score = outputText.length > 0 ? 1.0 : 0.0;
   return {
     pass: score >= 0.5,
     score,
@@ -3967,7 +3967,6 @@ var evalAssertCommand = command({
     }
     const payload = JSON.stringify(
       {
-        answer: resolvedOutput,
         output: [{ role: "assistant", content: resolvedOutput }],
         input: [{ role: "user", content: resolvedInput }],
         question: resolvedInput,
@@ -4185,7 +4184,7 @@ var evalRunCommand = command({
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-J7SUWZH2.js");
+      const { launchInteractiveWizard } = await import("./interactive-F6XECJ33.js");
       await launchInteractiveWizard();
       return;
     }
@@ -4707,10 +4706,10 @@ function exportResults(sourceFile, content, outputDir) {
   const outputsDir = path8.join(outputDir, "outputs");
   mkdirSync2(outputsDir, { recursive: true });
   for (const result of patched) {
-    const answer = result.answer;
-    if (answer) {
+    const outputText = result.outputText;
+    if (outputText) {
       const id = safeTestId(result);
-      writeFileSync3(path8.join(outputsDir, `${id}.txt`), answer);
+      writeFileSync3(path8.join(outputsDir, `${id}.txt`), outputText);
     }
   }
 }
@@ -5022,7 +5021,7 @@ function toTraceSummary(raw) {
   return toCamelCaseDeep(raw.trace);
 }
 function extractCandidate(raw) {
-  if (raw.answer !== void 0) return raw.answer;
+  if (raw.output_text !== void 0) return raw.output_text;
   if (raw.output !== void 0)
     return typeof raw.output === "string" ? raw.output : JSON.stringify(raw.output);
   return "";
@@ -5221,9 +5220,8 @@ var traceScoreCommand = command({
 function renderFlatTrace(result) {
   const trace = result.trace;
   const parts = [];
-  if (trace?.tool_names && trace.tool_names.length > 0) {
-    const toolParts = trace.tool_names.map((name) => {
-      const count = trace.tool_calls_by_name?.[name] ?? 0;
+  if (trace?.tool_calls && Object.keys(trace.tool_calls).length > 0) {
+    const toolParts = Object.entries(trace.tool_calls).map(([name, count]) => {
       return count > 1 ? `${name} \xD7${count}` : name;
     });
     parts.push(`Tools: ${toolParts.join(", ")}`);
@@ -6278,4 +6276,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-5GG6DDP5.js.map
+//# sourceMappingURL=chunk-TR6H437M.js.map