@agentv/core 3.9.0 → 3.9.2

package/dist/evaluation/validation/index.js

@@ -8,7 +8,7 @@ import {
   isEvaluatorKind,
   loadCasesFromFile,
   resolveFileReference
-} from "../../chunk-PC5TLJF6.js";
+} from "../../chunk-K7JCJIXA.js";
 
 // src/evaluation/validation/file-type.ts
 import { readFile } from "node:fs/promises";

package/dist/index.cjs

@@ -1575,6 +1575,7 @@ __export(index_exports, {
   extractTargetsFromSuite: () => extractTargetsFromSuite,
   extractTargetsFromTestCase: () => extractTargetsFromTestCase,
   extractTrialsConfig: () => extractTrialsConfig,
+  extractWorkersFromSuite: () => extractWorkersFromSuite,
   fileExists: () => fileExists2,
   findGitRoot: () => findGitRoot,
   freeformEvaluationSchema: () => freeformEvaluationSchema,
@@ -2256,6 +2257,17 @@ function extractTargetsFromSuite(suite) {
   }
   return void 0;
 }
+function extractWorkersFromSuite(suite) {
+  const execution = suite.execution;
+  if (!execution || typeof execution !== "object" || Array.isArray(execution)) {
+    return void 0;
+  }
+  const workers = execution.workers;
+  if (typeof workers === "number" && Number.isInteger(workers) && workers >= 1 && workers <= 50) {
+    return workers;
+  }
+  return void 0;
+}
 function extractTargetsFromTestCase(testCase) {
   const execution = testCase.execution;
   if (!execution || typeof execution !== "object" || Array.isArray(execution)) {
@@ -4556,6 +4568,7 @@ async function loadTestSuite(evalFilePath, repoRoot, options) {
     tests,
     trials: extractTrialsConfig(parsed),
     targets: extractTargetsFromSuite(parsed),
+    workers: extractWorkersFromSuite(parsed),
     cacheConfig: extractCacheConfig(parsed),
     totalBudgetUsd: extractTotalBudgetUsd(parsed),
     ...metadata !== void 0 && { metadata },
@@ -13344,8 +13357,7 @@ ${context2.fileChanges}`;
     }
     const evaluatorRawRequest = {
       userPrompt,
-      systemPrompt,
-      target: graderProvider.targetName
+      systemPrompt
     };
     try {
       const { data, tokenUsage } = await this.runWithRetry({
@@ -13363,6 +13375,7 @@ ${context2.fileChanges}`;
         assertions,
         expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
+        graderTarget: graderProvider.targetName,
         tokenUsage
       };
     } catch (e) {
@@ -13374,7 +13387,8 @@ ${context2.fileChanges}`;
         verdict: "skip",
         assertions: [{ text: `Grader parse failure after 3 attempts: ${message}`, passed: false }],
         expectedAspectCount: 1,
-        evaluatorRawRequest
+        evaluatorRawRequest,
+        graderTarget: graderProvider.targetName
       };
     }
   }
@@ -13392,8 +13406,7 @@ ${context2.fileChanges}`;
     const systemPrompt = buildRubricOutputSchema();
     const evaluatorRawRequest = {
       userPrompt: prompt,
-      systemPrompt,
-      target: graderProvider.targetName
+      systemPrompt
     };
     try {
       const { data, tokenUsage } = await this.runWithRetry({
@@ -13410,6 +13423,7 @@ ${context2.fileChanges}`;
         assertions,
         expectedAspectCount: rubrics.length,
         evaluatorRawRequest,
+        graderTarget: graderProvider.targetName,
         tokenUsage
       };
     } catch (e) {
@@ -13421,7 +13435,8 @@ ${context2.fileChanges}`;
         verdict: "skip",
         assertions: [{ text: `Grader parse failure after 3 attempts: ${message}`, passed: false }],
         expectedAspectCount: rubrics.length,
-        evaluatorRawRequest
+        evaluatorRawRequest,
+        graderTarget: graderProvider.targetName
       };
     }
   }
@@ -13434,8 +13449,7 @@ ${context2.fileChanges}`;
     const systemPrompt = buildScoreRangeOutputSchema();
     const evaluatorRawRequest = {
       userPrompt: prompt,
-      systemPrompt,
-      target: graderProvider.targetName
+      systemPrompt
     };
     try {
       const { data, tokenUsage } = await this.runWithRetry({
@@ -13452,6 +13466,7 @@ ${context2.fileChanges}`;
         assertions,
         expectedAspectCount: rubrics.length,
         evaluatorRawRequest,
+        graderTarget: graderProvider.targetName,
         details,
         tokenUsage
       };
@@ -13464,7 +13479,8 @@ ${context2.fileChanges}`;
         verdict: "skip",
         assertions: [{ text: `Grader parse failure after 3 attempts: ${message}`, passed: false }],
         expectedAspectCount: rubrics.length,
-        evaluatorRawRequest
+        evaluatorRawRequest,
+        graderTarget: graderProvider.targetName
       };
     }
   }
@@ -13496,7 +13512,6 @@ ${context2.fileChanges}`;
       mode: "built-in",
       systemPrompt,
       userPrompt,
-      target: graderProvider.targetName,
       maxSteps: this.maxSteps
     };
     try {
@@ -13514,7 +13529,13 @@ ${context2.fileChanges}`;
         steps: steps.length,
         tool_calls: toolCallCount
       };
-      return this.parseAgentResult(text, rubrics, evaluatorRawRequest, details);
+      return this.parseAgentResult(
+        text,
+        rubrics,
+        evaluatorRawRequest,
+        details,
+        graderProvider.targetName
+      );
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       return {
@@ -13523,6 +13544,7 @@ ${context2.fileChanges}`;
         assertions: [{ text: `llm-grader built-in evaluation failed: ${message}`, passed: false }],
         expectedAspectCount: 1,
         evaluatorRawRequest,
+        graderTarget: graderProvider.targetName,
         details: { mode: "built-in", error: message }
       };
     }
@@ -13575,6 +13597,7 @@ ${context2.fileChanges}`;
           ],
           expectedAspectCount: 1,
           evaluatorRawRequest,
+          graderTarget: provider.targetName,
           details: { mode: modeLabel, grader_target: provider.targetName }
         };
       }
@@ -13584,7 +13607,13 @@ ${context2.fileChanges}`;
         mode: modeLabel,
         grader_target: provider.targetName
       };
-      return this.parseAgentResult(assistantContent, rubrics, evaluatorRawRequest, details);
+      return this.parseAgentResult(
+        assistantContent,
+        rubrics,
+        evaluatorRawRequest,
+        details,
+        provider.targetName
+      );
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       return {
@@ -13595,6 +13624,7 @@ ${context2.fileChanges}`;
         ],
         expectedAspectCount: 1,
         evaluatorRawRequest,
+        graderTarget: provider.targetName,
         details: {
           mode: modeLabel,
           grader_target: provider.targetName,
@@ -13739,7 +13769,7 @@ ${outputSchema}`;
    * Parse the agent's response text into an EvaluationScore.
    * Supports both freeform and rubric modes.
    */
-  parseAgentResult(text, rubrics, evaluatorRawRequest, details) {
+  parseAgentResult(text, rubrics, evaluatorRawRequest, details, graderTarget) {
     try {
       const parsed = parseJsonFromText(text);
       if (rubrics && rubrics.length > 0) {
@@ -13751,6 +13781,7 @@ ${outputSchema}`;
           assertions: assertions2,
           expectedAspectCount: rubrics.length,
           evaluatorRawRequest,
+          graderTarget,
           details
         };
       }
@@ -13763,6 +13794,7 @@ ${outputSchema}`;
         assertions,
         expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
+        graderTarget,
         details
       };
     } catch {
@@ -13777,6 +13809,7 @@ ${outputSchema}`;
         ],
         expectedAspectCount: 1,
         evaluatorRawRequest,
+        graderTarget,
         details
       };
     }
@@ -17790,14 +17823,22 @@ async function runEvaluation(options) {
   const usePool = poolEnabled !== false && !!suiteWorkspace?.repos?.length && !isPerTestIsolation && !useStaticWorkspace;
   const resolvedRetainOnSuccess = retainOnSuccess ?? (keepWorkspaces ? "keep" : "cleanup");
   const resolvedRetainOnFailure = retainOnFailure ?? (cleanupWorkspaces ? "cleanup" : "keep");
-  const requestedWorkers = options.maxConcurrency ?? target.workers ?? 1;
-  const workers = hasSharedWorkspace && !usePool ? 1 : requestedWorkers;
+  const workers = options.maxConcurrency ?? target.workers ?? 1;
   setupLog(
-    `sharedWorkspace=${hasSharedWorkspace} perTestIsolation=${isPerTestIsolation} usePool=${usePool} requestedWorkers=${requestedWorkers} effectiveWorkers=${workers}`
+    `sharedWorkspace=${hasSharedWorkspace} perTestIsolation=${isPerTestIsolation} usePool=${usePool} workers=${workers}`
   );
-  if (hasSharedWorkspace && !usePool && requestedWorkers > 1) {
+  if (hasSharedWorkspace && !usePool && workers > 1) {
     console.warn(
-      `Warning: Shared workspace requires sequential execution. Overriding workers from ${requestedWorkers} to 1.`
+      [
+        `Warning: This eval uses a shared workspace with ${workers} workers.`,
+        "If the agent under test makes file edits, concurrent runs may corrupt each other.",
+        "To limit concurrency, add this to your eval YAML:",
+        "",
+        "  execution:",
+        "    workers: 1",
+        "",
+        "Or pass --workers 1 on the command line."
+      ].join("\n")
     );
   }
   const limit = pLimit(workers);
@@ -18080,7 +18121,8 @@ async function runEvaluation(options) {
             streamCallbacks,
             typeRegistry,
             repoManager,
-            evalDir
+            evalDir,
+            verbose
           };
           let result = trials && trials.count > 1 ? await runEvalCaseWithTrials(runCaseOptions, trials) : await runEvalCase(runCaseOptions);
           if (totalBudgetUsd !== void 0) {
@@ -18160,7 +18202,8 @@ async function runEvaluation(options) {
           promptInputs,
           primaryProvider,
           "agent",
-          "provider_error"
+          "provider_error",
+          verbose
         );
         results.push(errorResult);
         if (onResult) {
@@ -18233,6 +18276,7 @@ async function runBatchEvaluation(options) {
     nowFn,
     onProgress,
     onResult,
+    verbose,
     resolveGraderProvider,
     agentTimeoutMs,
     targetResolver,
@@ -18320,7 +18364,8 @@ async function runBatchEvaluation(options) {
         startTime,
         endTime,
         targetResolver,
-        availableTargets
+        availableTargets,
+        verbose
       });
       if (providerError) {
         result = {
@@ -18341,7 +18386,8 @@ async function runBatchEvaluation(options) {
         promptInputs,
         provider,
         "evaluator",
-        "evaluator_error"
+        "evaluator_error",
+        verbose
       );
       results.push(errorResult);
       if (onResult) {
@@ -18404,7 +18450,8 @@ async function runEvalCase(options) {
     suiteWorkspaceFile,
     typeRegistry: providedTypeRegistry,
     repoManager,
-    evalDir
+    evalDir,
+    verbose
   } = options;
   const setupDebug = process.env.AGENTV_SETUP_DEBUG === "1";
   const formattingMode = usesFileReferencePrompt(provider) ? "agent" : "lm";
@@ -18441,7 +18488,8 @@ async function runEvalCase(options) {
           promptInputs,
           provider,
           "setup",
-          "template_error"
+          "template_error",
+          verbose
         );
       }
       if (caseWorkspaceFile && workspacePath) {
@@ -18470,7 +18518,8 @@ async function runEvalCase(options) {
           promptInputs,
           provider,
           "repo_setup",
-          "local_path_not_found"
+          "local_path_not_found",
+          verbose
         );
       }
     }
@@ -18496,7 +18545,8 @@ async function runEvalCase(options) {
           promptInputs,
           provider,
           "repo_setup",
-          "clone_error"
+          "clone_error",
+          verbose
         );
       }
     }
@@ -18522,7 +18572,8 @@ async function runEvalCase(options) {
               promptInputs,
               provider,
               "setup",
-              "file_copy_error"
+              "file_copy_error",
+              verbose
             );
           }
         }
@@ -18567,7 +18618,8 @@ async function runEvalCase(options) {
           promptInputs,
           provider,
           "setup",
-          "script_error"
+          "script_error",
+          verbose
         );
       }
     }
@@ -18598,7 +18650,8 @@ async function runEvalCase(options) {
         promptInputs,
         provider,
         "setup",
-        "script_error"
+        "script_error",
+        verbose
       );
     }
   }
@@ -18642,7 +18695,8 @@ async function runEvalCase(options) {
         promptInputs,
         provider,
         "agent",
-        "provider_error"
+        "provider_error",
+        verbose
       );
       if (workspacePath) {
         if (forceCleanup) {
@@ -18663,7 +18717,8 @@ async function runEvalCase(options) {
       promptInputs,
       provider,
       "agent",
-      "provider_error"
+      "provider_error",
+      verbose
     );
     if (workspacePath) {
       if (forceCleanup) {
@@ -18758,7 +18813,8 @@ async function runEvalCase(options) {
       targetResolver,
       availableTargets,
       fileChanges,
-      workspacePath
+      workspacePath,
+      verbose
     });
     const totalDurationMs = Date.now() - caseStartMs;
     const graderTokens = aggregateEvaluatorTokenUsage(result.scores);
@@ -18813,7 +18869,8 @@ async function runEvalCase(options) {
       promptInputs,
       provider,
       "evaluator",
-      "evaluator_error"
+      "evaluator_error",
+      verbose
     );
     if (workspacePath && !isSharedWorkspace) {
       if (forceCleanup || (retainOnFailure ?? "keep") === "cleanup") {
@@ -18955,7 +19012,7 @@ async function evaluateCandidate(options) {
   let lmRequest;
   if (isAgentProvider(provider)) {
     agentRequest = {
-      question: promptInputs.question
+      ...options.verbose ? { input: promptInputs.question } : {}
     };
   } else {
     if (promptInputs.chatPrompt) {
@@ -18969,8 +19026,9 @@ async function evaluateCandidate(options) {
     }
   }
   const evaluatorRequest = scores ? void 0 : score.evaluatorRawRequest;
-  const requests = agentRequest || lmRequest || evaluatorRequest ? {
-    ...agentRequest ? { agent: agentRequest } : {},
+  const effectiveAgentRequest = agentRequest && Object.keys(agentRequest).length > 0 ? agentRequest : void 0;
+  const requests = effectiveAgentRequest || lmRequest || evaluatorRequest ? {
+    ...effectiveAgentRequest ? { agent: effectiveAgentRequest } : {},
     ...lmRequest ? { lm: lmRequest } : {},
     ...evaluatorRequest ? { evaluator: evaluatorRequest } : {}
   } : void 0;
@@ -18990,9 +19048,9 @@ async function evaluateCandidate(options) {
     endTime,
     requests,
     input,
+    output: output ?? [{ role: "assistant", content: candidate }],
     scores,
     trace: trace2,
-    output: output ?? [{ role: "assistant", content: candidate }],
     fileChanges,
     executionStatus: classifyQualityStatus(score.score)
   };
@@ -19158,6 +19216,7 @@ async function runEvaluatorList(options) {
         verdict: score2.verdict,
         assertions: score2.assertions,
         input: score2.evaluatorRawRequest,
+        target: score2.graderTarget,
         details: score2.details,
         scores: mapChildResults(score2.scores),
         tokenUsage: score2.tokenUsage,
@@ -19297,13 +19356,13 @@ async function invokeProvider(provider, options) {
     }
   }
 }
-function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs, provider, failureStage, failureReasonCode) {
+function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs, provider, failureStage, failureReasonCode, verbose) {
   const message = error instanceof Error ? error.message : String(error);
   let agentRequest;
   let lmRequest;
   if (isAgentProvider(provider)) {
     agentRequest = {
-      question: promptInputs.question,
+      ...verbose ? { input: promptInputs.question } : {},
       error: message
     };
   } else {
@@ -19331,10 +19390,10 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     conversationId: evalCase.conversation_id,
     score: 0,
     assertions: [{ text: `Error: ${message}`, passed: false }],
-    output: [{ role: "assistant", content: `Error occurred: ${message}` }],
     target: targetName,
     requests,
     input,
+    output: [{ role: "assistant", content: `Error occurred: ${message}` }],
     error: message,
     executionStatus: "execution_error",
     failureStage,
@@ -20408,6 +20467,7 @@ function createAgentKernel() {
   extractTargetsFromSuite,
   extractTargetsFromTestCase,
   extractTrialsConfig,
+  extractWorkersFromSuite,
   fileExists,
   findGitRoot,
   freeformEvaluationSchema,