agentv 3.4.0 → 3.6.0

package/dist/{chunk-A7ZDUB46.js → chunk-IP5BO54H.js}

@@ -16,7 +16,7 @@ import {
   validateEvalFile,
   validateFileReferences,
   validateTargetsFile
-} from "./chunk-RE5I3U2S.js";
+} from "./chunk-UU5N43YS.js";
 import {
   createBuiltinRegistry,
   createProvider,
@@ -34,7 +34,7 @@ import {
   toSnakeCaseDeep as toSnakeCaseDeep2,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-GOZV2HN2.js";
+} from "./chunk-K4RXLQWV.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-WXXTZ7PD.js");
+      const { launchInteractiveWizard } = await import("./interactive-5S4ILY2Y.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 "";
@@ -5089,9 +5088,7 @@ async function runScore(results, evaluatorConfig, testIdFilter) {
       originalScore: raw.score,
       newScore: score.score,
       verdict: score.verdict,
-      hits: score.hits,
-      misses: score.misses,
-      reasoning: score.reasoning
+      assertions: score.assertions
     });
   }
   return scored;
@@ -5110,7 +5107,9 @@ function renderTable(scored, assertSpec) {
   lines.push(cols.map((col) => "\u2500".repeat(col.width)).join("\u2500\u2500"));
   for (const r of scored) {
     const verdictColor = r.verdict === "pass" ? c2.green : c2.red;
-    const detail = r.misses.length > 0 ? r.misses[0].slice(0, 48) : r.hits.length > 0 ? r.hits[0].slice(0, 48) : r.reasoning?.slice(0, 48) ?? "";
+    const failed = r.assertions.filter((a) => !a.passed);
+    const passed = r.assertions.filter((a) => a.passed);
+    const detail = failed.length > 0 ? failed[0].text.slice(0, 48) : passed.length > 0 ? passed[0].text.slice(0, 48) : "";
     const row = [
       padRight2(r.testId.slice(0, 24), cols[0].width),
       padLeft2(formatScore(r.originalScore), cols[1].width),
@@ -5332,11 +5331,17 @@ function formatResultDetail(result, index, tree) {
   if (result.error) {
     lines.push(`  ${c2.red}Error: ${result.error}${c2.reset}`);
   }
-  if (result.hits && result.hits.length > 0) {
-    lines.push(`  ${c2.green}\u2713 Hits:${c2.reset} ${result.hits.join(", ")}`);
-  }
-  if (result.misses && result.misses.length > 0) {
-    lines.push(`  ${c2.red}\u2717 Misses:${c2.reset} ${result.misses.join(", ")}`);
+  if (result.assertions && result.assertions.length > 0) {
+    const passed = result.assertions.filter((a) => a.passed);
+    const failed = result.assertions.filter((a) => !a.passed);
+    if (passed.length > 0)
+      lines.push(
+        `  ${c2.green}\u2713 Passed:${c2.reset} ${passed.map((a) => a.text).join(", ")}`
+      );
+    if (failed.length > 0)
+      lines.push(
+        `  ${c2.red}\u2717 Failed:${c2.reset} ${failed.map((a) => a.text).join(", ")}`
+      );
   }
   if (result.scores && result.scores.length > 0) {
     lines.push(`  ${c2.dim}Scores:${c2.reset} ${renderScores(result.scores)}`);
@@ -5344,10 +5349,14 @@ function formatResultDetail(result, index, tree) {
   if (result.trace || result.duration_ms !== void 0 || result.cost_usd !== void 0) {
     lines.push(`  ${c2.dim}Trace:${c2.reset} ${renderFlatTrace(result)}`);
   }
-  if (result.reasoning) {
-    const maxLen = 200;
-    const truncated = result.reasoning.length > maxLen ? `${result.reasoning.slice(0, maxLen)}...` : result.reasoning;
-    lines.push(`  ${c2.dim}Reasoning: ${truncated}${c2.reset}`);
+  if (result.assertions && result.assertions.length > 0) {
+    const withEvidence = result.assertions.filter((a) => a.evidence);
+    if (withEvidence.length > 0) {
+      const maxLen = 200;
+      const evidence = withEvidence[0].evidence;
+      const truncated = evidence.length > maxLen ? `${evidence.slice(0, maxLen)}...` : evidence;
+      lines.push(`  ${c2.dim}Evidence: ${truncated}${c2.reset}`);
+    }
   }
   return lines.join("\n");
 }
@@ -6268,4 +6277,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-A7ZDUB46.js.map
+//# sourceMappingURL=chunk-IP5BO54H.js.map