agentv 3.9.1 → 3.10.0

package/dist/{chunk-FRA6PDLZ.js → chunk-DDMAQT5P.js}

@@ -27,12 +27,12 @@ import {
   subscribeToCopilotCliLogEntries,
   subscribeToCopilotSdkLogEntries,
   subscribeToPiLogEntries
-} from "./chunk-X24J6HCV.js";
+} from "./chunk-X3KJVUAB.js";
 
 // package.json
 var package_default = {
   name: "agentv",
-  version: "3.9.1",
+  version: "3.10.0",
   description: "CLI entry point for AgentV",
   type: "module",
   repository: {
@@ -204,7 +204,7 @@ async function discoverTargetsFile(options) {
 // src/commands/eval/run-eval.ts
 import { constants as constants4 } from "node:fs";
 import { access as access4 } from "node:fs/promises";
-import path12 from "node:path";
+import path13 from "node:path";
 import { pathToFileURL } from "node:url";
 
 // src/version-check.ts
@@ -282,7 +282,7 @@ function computePassRate(result) {
     const passed = scores.filter((s) => s.score >= PASS_THRESHOLD).length;
     return passed / scores.length;
   }
-  return result.score >= PASS_THRESHOLD ? 1 : 0;
+  return (result.score ?? 0) >= PASS_THRESHOLD ? 1 : 0;
 }
 function countToolCalls(result) {
   const toolCalls = {};
@@ -323,6 +323,7 @@ function parseWorkspaceChanges(fileChanges) {
   };
 }
 function buildAssertions(result) {
+  if (!result.assertions) return [];
   return result.assertions.map((a) => ({
     text: a.text,
     passed: a.passed,
@@ -403,8 +404,8 @@ function buildBenchmarkArtifact(results, evalFile = "") {
   const targetSet = /* @__PURE__ */ new Set();
   const testIdSet = /* @__PURE__ */ new Set();
   for (const result of results) {
-    targetSet.add(result.target);
-    testIdSet.add(result.testId);
+    targetSet.add(result.target ?? "unknown");
+    testIdSet.add(result.testId ?? "unknown");
   }
   const targets = [...targetSet].sort();
   const testIds = [...testIdSet].sort();
@@ -452,7 +453,9 @@ function buildBenchmarkArtifact(results, evalFile = "") {
       perEvaluatorSummary[key] = computeStats(scores);
     }
   }
-  const errorCount = results.filter((r) => r.executionStatus === "execution_error").length;
+  const errorCount = results.filter(
+    (r) => r.executionStatus != null && r.executionStatus === "execution_error"
+  ).length;
   if (errorCount > 0) {
     notes.push(
       `${errorCount} test(s) had execution errors and are included in pass_rate as failures`
@@ -518,7 +521,7 @@ async function writeArtifactsFromResults(results, outputDir, options) {
   await mkdir(gradingDir, { recursive: true });
   for (const result of results) {
     const grading = buildGradingArtifact(result);
-    const safeTestId = result.testId.replace(/[/\\:*?"<>|]/g, "_");
+    const safeTestId = (result.testId ?? "unknown").replace(/[/\\:*?"<>|]/g, "_");
     const gradingPath = path3.join(gradingDir, `${safeTestId}.json`);
     await writeFile(gradingPath, `${JSON.stringify(grading, null, 2)}
 `, "utf8");
@@ -1766,12 +1769,12 @@ var ProgressDisplay = class {
   }
   addLogPaths(paths, provider) {
     const newPaths = [];
-    for (const path13 of paths) {
-      if (this.logPathSet.has(path13)) {
+    for (const path14 of paths) {
+      if (this.logPathSet.has(path14)) {
         continue;
       }
-      this.logPathSet.add(path13);
-      newPaths.push(path13);
+      this.logPathSet.add(path14);
+      newPaths.push(path14);
     }
     if (newPaths.length === 0) {
       return;
@@ -1784,8 +1787,8 @@ var ProgressDisplay = class {
       this.hasPrintedLogHeader = true;
     }
     const startIndex = this.logPaths.length - newPaths.length;
-    newPaths.forEach((path13, offset) => {
-      console.log(`${startIndex + offset + 1}. ${path13}`);
+    newPaths.forEach((path14, offset) => {
+      console.log(`${startIndex + offset + 1}. ${path14}`);
     });
   }
   finish() {
@@ -1859,6 +1862,32 @@ async function loadNonErrorResults(jsonlPath) {
   return results;
 }
 
+// src/commands/eval/run-cache.ts
+import { mkdir as mkdir7, readFile as readFile2, writeFile as writeFile6 } from "node:fs/promises";
+import path11 from "node:path";
+var CACHE_FILENAME = "cache.json";
+function cachePath(cwd) {
+  return path11.join(cwd, ".agentv", CACHE_FILENAME);
+}
+async function loadRunCache(cwd) {
+  try {
+    const content = await readFile2(cachePath(cwd), "utf-8");
+    return JSON.parse(content);
+  } catch {
+    return void 0;
+  }
+}
+async function saveRunCache(cwd, resultFile) {
+  const dir = path11.join(cwd, ".agentv");
+  await mkdir7(dir, { recursive: true });
+  const cache = {
+    lastResultFile: resultFile,
+    timestamp: (/* @__PURE__ */ new Date()).toISOString()
+  };
+  await writeFile6(cachePath(cwd), `${JSON.stringify(cache, null, 2)}
+`, "utf-8");
+}
+
 // src/commands/eval/statistics.ts
 var HISTOGRAM_BREAKPOINTS = [0, 0.2, 0.4, 0.6, 0.8, 1];
 function computeMean(values) {
@@ -2107,13 +2136,13 @@ function formatMatrixSummary(results) {
 }
 
 // ../../packages/core/dist/evaluation/validation/index.js
-import { readFile as readFile2 } from "node:fs/promises";
-import path11 from "node:path";
+import { readFile as readFile3 } from "node:fs/promises";
+import path12 from "node:path";
 import { parse } from "yaml";
 import { readFile as readFile22 } from "node:fs/promises";
 import path22 from "node:path";
 import { parse as parse2 } from "yaml";
-import { readFile as readFile3 } from "node:fs/promises";
+import { readFile as readFile32 } from "node:fs/promises";
 import path32 from "node:path";
 import { parse as parse3 } from "yaml";
 import { readFile as readFile4 } from "node:fs/promises";
@@ -2126,7 +2155,7 @@ var SCHEMA_TARGETS_V2 = "agentv-targets-v2.2";
 var SCHEMA_CONFIG_V2 = "agentv-config-v2";
 async function detectFileType(filePath) {
   try {
-    const content = await readFile2(filePath, "utf8");
+    const content = await readFile3(filePath, "utf8");
     const parsed = parse(content);
     if (typeof parsed !== "object" || parsed === null) {
       return inferFileTypeFromPath(filePath);
@@ -2151,8 +2180,8 @@ async function detectFileType(filePath) {
   }
 }
 function inferFileTypeFromPath(filePath) {
-  const normalized = path11.normalize(filePath).replace(/\\/g, "/");
-  const basename = path11.basename(filePath);
+  const normalized = path12.normalize(filePath).replace(/\\/g, "/");
+  const basename = path12.basename(filePath);
   if (normalized.includes("/.agentv/")) {
     if (basename === "config.yaml" || basename === "config.yml") {
       return "config";
@@ -3008,8 +3037,8 @@ async function validateTargetsFile(filePath) {
   const absolutePath = path32.resolve(filePath);
   let parsed;
   try {
-    const content = await readFile3(absolutePath, "utf8");
-    parsed = parse3(content);
+    const content = await readFile32(absolutePath, "utf8");
+    parsed = interpolateEnv(parse3(content), process.env);
   } catch (error) {
     errors.push({
       severity: "error",
@@ -3784,7 +3813,7 @@ function buildDefaultOutputPath(cwd, format) {
   const timestamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
   const baseName = "eval";
   const extension = getDefaultExtension(format);
-  return path12.join(cwd, ".agentv", "results", `${baseName}_${timestamp}${extension}`);
+  return path13.join(cwd, ".agentv", "results", `${baseName}_${timestamp}${extension}`);
 }
 function createProgressReporter(maxWorkers, options) {
   const display = new ProgressDisplay(maxWorkers, options);
@@ -3798,7 +3827,7 @@ function createProgressReporter(maxWorkers, options) {
   };
 }
 function makeEvalKey(testFilePath, evalId) {
-  return `${path12.resolve(testFilePath)}::${evalId}`;
+  return `${path13.resolve(testFilePath)}::${evalId}`;
 }
 function createDisplayIdTracker() {
   const map = /* @__PURE__ */ new Map();
@@ -4080,7 +4109,7 @@ async function runEvalCommand(input) {
     );
   }
   const repoRoot = await findRepoRoot(cwd);
-  const yamlConfig = await loadConfig(path12.join(cwd, "_"), repoRoot);
+  const yamlConfig = await loadConfig(path13.join(cwd, "_"), repoRoot);
   if (yamlConfig?.required_version) {
     await enforceRequiredVersion(yamlConfig.required_version, {
       strict: normalizeBoolean(input.rawOptions.strict)
@@ -4092,7 +4121,7 @@ async function runEvalCommand(input) {
   }
   let retryNonErrorResults;
   if (options.retryErrors) {
-    const retryPath = path12.resolve(options.retryErrors);
+    const retryPath = path13.resolve(options.retryErrors);
     await ensureFileExists(retryPath, "Retry-errors JSONL file");
     const errorIds = await loadErrorTestIds(retryPath);
     if (errorIds.length === 0) {
@@ -4105,7 +4134,7 @@ async function runEvalCommand(input) {
     retryNonErrorResults = await loadNonErrorResults(retryPath);
   }
   if (options.workspacePath) {
-    const resolvedWorkspace = path12.resolve(options.workspacePath);
+    const resolvedWorkspace = path13.resolve(options.workspacePath);
     try {
       const { stat: stat2 } = await import("node:fs/promises");
       const stats = await stat2(resolvedWorkspace);
@@ -4127,7 +4156,7 @@ async function runEvalCommand(input) {
   const useFileExport = !!(options.otelFile || options.traceFile);
   if (options.exportOtel || useFileExport) {
     try {
-      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-LPIGPS52.js");
+      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-DJFWBJNJ.js");
       let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
       let headers = {};
       if (options.otelBackend) {
@@ -4151,8 +4180,8 @@ async function runEvalCommand(input) {
         headers,
         captureContent,
         groupTurns: options.otelGroupTurns,
-        otlpFilePath: options.otelFile ? path12.resolve(options.otelFile) : void 0,
-        traceFilePath: options.traceFile ? path12.resolve(options.traceFile) : void 0
+        otlpFilePath: options.otelFile ? path13.resolve(options.otelFile) : void 0,
+        traceFilePath: options.traceFile ? path13.resolve(options.traceFile) : void 0
       });
       const initialized = await otelExporter.init();
       if (!initialized) {
@@ -4168,8 +4197,8 @@ async function runEvalCommand(input) {
       otelExporter = null;
     }
   }
-  const outputPath = options.outPath ? path12.resolve(options.outPath) : buildDefaultOutputPath(cwd, options.format);
-  const extraOutputPaths = options.outputPaths.map((p) => path12.resolve(p));
+  const outputPath = options.outPath ? path13.resolve(options.outPath) : buildDefaultOutputPath(cwd, options.format);
+  const extraOutputPaths = options.outputPaths.map((p) => path13.resolve(p));
   const allOutputPaths = extraOutputPaths.length > 0 ? [outputPath, ...extraOutputPaths] : [outputPath];
   const uniqueOutputPaths = [...new Set(allOutputPaths)];
   let outputWriter;
@@ -4183,12 +4212,12 @@ async function runEvalCommand(input) {
       console.log(`  ${p}`);
     }
   }
-  const resolvedTestFiles = input.testFiles.map((file) => path12.resolve(file));
+  const resolvedTestFiles = input.testFiles.map((file) => path13.resolve(file));
   if (options.otelFile) {
-    console.log(`OTLP JSON file: ${path12.resolve(options.otelFile)}`);
+    console.log(`OTLP JSON file: ${path13.resolve(options.otelFile)}`);
   }
   if (options.traceFile) {
-    console.log(`Trace file: ${path12.resolve(options.traceFile)}`);
+    console.log(`Trace file: ${path13.resolve(options.traceFile)}`);
   }
   const evaluationRunner = await resolveEvaluationRunner();
   const allResults = [];
@@ -4234,7 +4263,7 @@ async function runEvalCommand(input) {
     cliNoCache: options.noCache,
     yamlCache: yamlCacheEnabled
   });
-  const cache = cacheEnabled ? new ResponseCache(yamlCachePath ? path12.resolve(yamlCachePath) : void 0) : void 0;
+  const cache = cacheEnabled ? new ResponseCache(yamlCachePath ? path13.resolve(yamlCachePath) : void 0) : void 0;
   const useCache = cacheEnabled;
   if (cacheEnabled) {
     console.log(`Response cache: enabled${yamlCachePath ? ` (${yamlCachePath})` : ""}`);
@@ -4366,12 +4395,12 @@ async function runEvalCommand(input) {
       console.log(formatMatrixSummary(allResults));
     }
     if (options.benchmarkJson && allResults.length > 0) {
-      const benchmarkPath = path12.resolve(options.benchmarkJson);
+      const benchmarkPath = path13.resolve(options.benchmarkJson);
       await writeBenchmarkJson(benchmarkPath, allResults);
       console.log(`Benchmark written to: ${benchmarkPath}`);
     }
     if (options.artifacts && allResults.length > 0) {
-      const artifactsDir = path12.resolve(options.artifacts);
+      const artifactsDir = path13.resolve(options.artifacts);
       const evalFile = resolvedTestFiles.length === 1 ? resolvedTestFiles[0] : "";
       const {
         gradingDir,
@@ -4402,11 +4431,12 @@ Results written to: ${outputPath}`);
           console.log(`  ${p}`);
         }
       }
+      await saveRunCache(cwd, outputPath).catch(() => void 0);
     }
     if (summary.executionErrorCount > 0 && !options.retryErrors) {
-      const evalFileArgs = resolvedTestFiles.map((f) => path12.relative(cwd, f)).join(" ");
+      const evalFileArgs = resolvedTestFiles.map((f) => path13.relative(cwd, f)).join(" ");
       const targetFlag = options.target ? ` --target ${options.target}` : "";
-      const relativeOutputPath = path12.relative(cwd, outputPath);
+      const relativeOutputPath = path13.relative(cwd, outputPath);
       console.log(
         `
 Tip: ${summary.executionErrorCount} execution error(s) detected. Re-run failed tests with:
@@ -4438,7 +4468,7 @@ async function resolveEvaluationRunner() {
   if (!overridePath) {
     return runEvaluation;
   }
-  const resolved = path12.isAbsolute(overridePath) ? overridePath : path12.resolve(process.cwd(), overridePath);
+  const resolved = path13.isAbsolute(overridePath) ? overridePath : path13.resolve(process.cwd(), overridePath);
   const moduleUrl = pathToFileURL(resolved).href;
   const mod = await import(moduleUrl);
   const candidate = mod.runEvaluation;
@@ -4460,6 +4490,7 @@ export {
   buildTimingArtifact,
   buildBenchmarkArtifact,
   parseJsonlResults,
+  loadRunCache,
   detectFileType,
   validateEvalFile,
   validateTargetsFile,
@@ -4470,4 +4501,4 @@ export {
   selectTarget,
   runEvalCommand
 };
-//# sourceMappingURL=chunk-FRA6PDLZ.js.map
+//# sourceMappingURL=chunk-DDMAQT5P.js.map