agentv 4.18.0-next.1 → 4.19.0

package/dist/{chunk-VRPCMCLQ.js → chunk-IWI4AJRS.js}

@@ -15,7 +15,22 @@ import {
   resolveWorkspaceOrFilePath,
   toSnakeCaseDeep,
   writeArtifactsFromResults
-} from "./chunk-HBDOJJFY.js";
+} from "./chunk-NL6P5MUH.js";
+import {
+  ResponseCache,
+  commitAndPushResultsBranch,
+  createDraftResultsPr,
+  deriveCategory,
+  directorySizeBytes,
+  getResultsRepoStatus,
+  loadTsConfig,
+  prepareResultsRepoBranch,
+  resolveResultsRepoRunsDir,
+  shouldEnableCache,
+  shouldSkipCacheForTemperature,
+  stageResultsArtifacts,
+  syncResultsRepo
+} from "./chunk-R2QDYORI.js";
 import {
   CLI_PLACEHOLDERS,
   COMMON_TARGET_SETTINGS,
@@ -23,47 +38,34 @@ import {
   DEFAULT_THRESHOLD,
   KNOWN_PROVIDERS,
   PROVIDER_ALIASES,
-  ResponseCache,
   buildDirectoryChain,
   buildSearchRoots,
-  commitAndPushResultsBranch,
-  createDraftResultsPr,
-  deriveCategory,
-  directorySizeBytes,
   ensureVSCodeSubagents,
   findDeprecatedCamelCaseTargetWarnings,
   findGitRoot,
-  getResultsRepoStatus,
   interpolateEnv,
   isGraderKind,
   listTargetNames,
   loadCasesFromFile,
   loadConfig,
   loadTestSuite,
-  loadTsConfig,
   normalizeLineEndings,
-  prepareResultsRepoBranch,
   readTargetDefinitions,
   readTestSuiteMetadata,
   resolveFileReference,
-  resolveResultsRepoRunsDir,
   resolveTargetDefinition,
   runEvaluation,
-  shouldEnableCache,
-  shouldSkipCacheForTemperature,
-  stageResultsArtifacts,
   subscribeToCodexLogEntries,
   subscribeToCopilotCliLogEntries,
   subscribeToCopilotSdkLogEntries,
   subscribeToPiLogEntries,
-  syncResultsRepo,
   toCamelCaseDeep
-} from "./chunk-RCOAXXHP.js";
+} from "./chunk-PTYQS37Y.js";
 
 // package.json
 var package_default = {
   name: "agentv",
-  version: "4.18.0-next.1",
+  version: "4.19.0",
   description: "CLI entry point for AgentV",
   type: "module",
   repository: {
@@ -148,12 +150,15 @@ async function resolveEvalPaths(evalPaths, cwd) {
     const candidatePath = path.isAbsolute(pattern) ? path.normalize(pattern) : path.resolve(cwd, pattern);
     try {
       const stats = await stat(candidatePath);
-      if (stats.isFile() && /\.(ya?ml|jsonl|json)$/i.test(candidatePath)) {
+      if (stats.isFile() && /\.(ya?ml|jsonl|json|[cm]?ts)$/i.test(candidatePath)) {
         results.add(candidatePath);
         continue;
       }
       if (stats.isDirectory()) {
-        const dirGlob = path.posix.join(candidatePath.replace(/\\/g, "/"), "**/*.eval.{yaml,yml}");
+        const dirGlob = path.posix.join(
+          candidatePath.replace(/\\/g, "/"),
+          "**/{*.eval.yaml,*.eval.yml,eval.yaml,eval.yml,*.eval.ts,*.eval.mts}"
+        );
         const dirMatches = await fg(dirGlob, {
           absolute: true,
           onlyFiles: true,
@@ -179,7 +184,9 @@ async function resolveEvalPaths(evalPaths, cwd) {
       followSymbolicLinks: true,
       ignore: ignorePatterns
     });
-    const yamlMatches = matches.filter((filePath) => /\.(ya?ml|jsonl|json)$/i.test(filePath));
+    const yamlMatches = matches.filter(
+      (filePath) => /\.(ya?ml|jsonl|json|[cm]?ts)$/i.test(filePath)
+    );
     for (const filePath of yamlMatches) {
       results.add(path.normalize(filePath));
     }
@@ -201,7 +208,7 @@ async function resolveEvalPaths(evalPaths, cwd) {
     throw new Error(
       `No eval files matched any provided paths or globs: ${includePatterns.join(
         ", "
-      )}. Provide YAML, JSONL, or JSON paths or globs (e.g., "evals/**/*.yaml", "evals/**/*.jsonl", "evals.json").`
+      )}. Provide YAML, JSONL, JSON, or TypeScript paths or globs (e.g., "evals/**/eval.yaml", "evals/**/*.eval.ts").`
     );
   }
   const sorted = Array.from(results);
@@ -4874,6 +4881,52 @@ async function prepareFileMetadata(params) {
         inlineTargetLabel: `transcript (${path15.basename(options.transcript)})`
       }
     ];
+  } else if (suite.inlineTarget && options.cliTargets.length === 0) {
+    const targetDefinition = suite.inlineTarget;
+    const resolvedTarget = options.dryRun ? {
+      kind: "mock",
+      name: `${targetDefinition.name}-dry-run`,
+      graderTarget: void 0,
+      config: {
+        response: '{"answer":"Mock dry-run response"}',
+        delayMs: options.dryRunDelay,
+        delayMinMs: options.dryRunDelayMin,
+        delayMaxMs: options.dryRunDelayMax
+      }
+    } : resolveTargetDefinition(targetDefinition, process.env, testFilePath, {
+      emitDeprecationWarnings: false
+    });
+    selections = [
+      {
+        selection: {
+          definitions: [targetDefinition],
+          resolvedTarget,
+          targetName: targetDefinition.name,
+          targetSource: "test-file",
+          targetsFilePath: testFilePath
+        },
+        inlineTargetLabel: resolveTargetLabel(targetDefinition.name, resolvedTarget.name)
+      }
+    ];
+  } else if (suite.providerFactory && options.cliTargets.length === 0) {
+    const taskTarget = {
+      kind: "mock",
+      name: "custom-task",
+      graderTarget: void 0,
+      config: {}
+    };
+    selections = [
+      {
+        selection: {
+          definitions: [],
+          resolvedTarget: taskTarget,
+          targetName: "custom-task",
+          targetSource: "test-file",
+          targetsFilePath: testFilePath
+        },
+        inlineTargetLabel: "custom-task"
+      }
+    ];
   } else {
     const cliTargets = options.cliTargets;
     const suiteTargets2 = suite.targets;
@@ -4944,7 +4997,8 @@ async function prepareFileMetadata(params) {
     budgetUsd: suite.budgetUsd,
     failOnError: suite.failOnError,
     threshold: suite.threshold,
-    tags: suite.metadata?.tags
+    tags: suite.metadata?.tags,
+    providerFactory: suite.providerFactory
   };
 }
 async function runSingleEvalFile(params) {
@@ -5223,7 +5277,7 @@ async function runEvalCommand(input) {
   const useFileExport = !!options.otelFile;
   if (options.exportOtel || useFileExport) {
     try {
-      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-7W4OI3X2.js");
+      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-RTIUSC6L.js");
       let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
       let headers = {};
       if (options.otelBackend) {
@@ -5283,23 +5337,7 @@ async function runEvalCommand(input) {
   const displayIdTracker = createDisplayIdTracker();
   const perFileWorkers = options.workers;
   const fileMetadata = /* @__PURE__ */ new Map();
-  const tsFiles = [];
-  const yamlFiles = [];
   for (const testFilePath of resolvedTestFiles) {
-    if (/\.(ts|js|mts|mjs)$/.test(testFilePath)) {
-      tsFiles.push(testFilePath);
-    } else {
-      yamlFiles.push(testFilePath);
-    }
-  }
-  for (const tsFile of tsFiles) {
-    await ensureFileExists(tsFile, "TypeScript eval file");
-    await import(pathToFileURL(tsFile).href);
-  }
-  if (yamlFiles.length === 0 && tsFiles.length > 0) {
-    return;
-  }
-  for (const testFilePath of yamlFiles) {
     const meta = await prepareFileMetadata({
       testFilePath,
       repoRoot,
@@ -5434,7 +5472,7 @@ async function runEvalCommand(input) {
   const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
   let transcriptProviderFactory;
   if (options.transcript) {
-    const { TranscriptProvider } = await import("./dist-7W4OI3X2.js");
+    const { TranscriptProvider } = await import("./dist-RTIUSC6L.js");
     const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
     const totalTests = [...fileMetadata.values()].reduce(
       (sum, meta) => sum + meta.testCases.length,
@@ -5494,7 +5532,7 @@ async function runEvalCommand(input) {
               budgetUsd: targetPrep.budgetUsd,
               failOnError: targetPrep.failOnError,
               threshold: resolvedThreshold,
-              providerFactory: transcriptProviderFactory
+              providerFactory: transcriptProviderFactory ?? targetPrep.providerFactory
             });
             const evalFile = path15.relative(cwd, testFilePath);
             const existingSummary = remoteEvalSummaries.find(
@@ -5574,7 +5612,7 @@ async function runEvalCommand(input) {
     if (usesDefaultArtifactWorkspace && allResults.length > 0) {
       const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
       if (isResumeAppend) {
-        const { writePerTestArtifacts } = await import("./artifact-writer-WH3OE42V.js");
+        const { writePerTestArtifacts } = await import("./artifact-writer-YATMDPWI.js");
         await writePerTestArtifacts(allResults, runDir, {
           experiment: normalizeExperimentName(options.experiment)
         });
@@ -5784,4 +5822,4 @@ export {
   getCategories,
   filterByCategory
 };
-//# sourceMappingURL=chunk-VRPCMCLQ.js.map
+//# sourceMappingURL=chunk-IWI4AJRS.js.map