agentv 3.4.0 → 3.5.0

package/README.md

@@ -238,21 +238,19 @@ import json, sys
 data = json.load(sys.stdin)
 answer = data.get("answer", "")
 
-hits = []
-misses = []
+assertions = []
 
 if "42" in answer:
-    hits.append("Answer contains correct value (42)")
+    assertions.append({"text": "Answer contains correct value (42)", "passed": True})
 else:
-    misses.append("Answer does not contain expected value (42)")
+    assertions.append({"text": "Answer does not contain expected value (42)", "passed": False})
 
-score = 1.0 if hits else 0.0
+passed = sum(1 for a in assertions if a["passed"])
+score = 1.0 if passed == len(assertions) else 0.0
 
 print(json.dumps({
     "score": score,
-    "hits": hits,
-    "misses": misses,
-    "reasoning": f"Passed {len(hits)} check(s)"
+    "assertions": assertions,
 }))
 ```
 

package/dist/{agentv-provider-HDSAUUEF-LUBMM7TH.js → agentv-provider-NFFLXG5M-TJAWCWCX.js}

@@ -1,9 +1,9 @@
 import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
 import {
   AgentvProvider
-} from "./chunk-AR3QEKXH.js";
+} from "./chunk-BJV6MDBE.js";
 import "./chunk-5H446C7X.js";
 export {
   AgentvProvider
 };
-//# sourceMappingURL=agentv-provider-HDSAUUEF-LUBMM7TH.js.map
+//# sourceMappingURL=agentv-provider-NFFLXG5M-TJAWCWCX.js.map

package/dist/{chunk-A7ZDUB46.js → chunk-5GG6DDP5.js}

@@ -16,7 +16,7 @@ import {
   validateEvalFile,
   validateFileReferences,
   validateTargetsFile
-} from "./chunk-RE5I3U2S.js";
+} from "./chunk-RLL4QGNL.js";
 import {
   createBuiltinRegistry,
   createProvider,
@@ -34,7 +34,7 @@ import {
   toSnakeCaseDeep as toSnakeCaseDeep2,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-GOZV2HN2.js";
+} from "./chunk-D6G4N2H2.js";
 import {
   __commonJS,
   __esm,
@@ -4185,7 +4185,7 @@ 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-J7SUWZH2.js");
       await launchInteractiveWizard();
       return;
     }
@@ -5089,9 +5089,7 @@ 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 +5108,9 @@ 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 +5332,17 @@ 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 +5350,14 @@ 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 +6278,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-A7ZDUB46.js.map
+//# sourceMappingURL=chunk-5GG6DDP5.js.map