agentv 4.20.0 → 4.21.0

package/dist/{chunk-ZNS74WKH.js → chunk-FZUNMXBT.js}

@@ -1,10 +1,10 @@
 import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
 import {
   toTranscriptJsonLines
-} from "./chunk-36HXBYUY.js";
+} from "./chunk-27VT2KU2.js";
 import {
   DEFAULT_THRESHOLD
-} from "./chunk-LP4Y5D2Z.js";
+} from "./chunk-ERSBQAGK.js";
 
 // src/commands/eval/artifact-writer.ts
 import { mkdir, readFile, writeFile } from "node:fs/promises";
@@ -688,4 +688,4 @@ export {
   writePerTestArtifacts,
   writeArtifactsFromResults
 };
-//# sourceMappingURL=chunk-ZNS74WKH.js.map
+//# sourceMappingURL=chunk-FZUNMXBT.js.map

package/dist/{chunk-PHGEGHKR.js → chunk-GPGX24OI.js}

@@ -15,7 +15,7 @@ import {
   resolveWorkspaceOrFilePath,
   toSnakeCaseDeep,
   writeArtifactsFromResults
-} from "./chunk-ZNS74WKH.js";
+} from "./chunk-FZUNMXBT.js";
 import {
   ResponseCache,
   RunBudgetTracker,
@@ -31,7 +31,7 @@ import {
   shouldSkipCacheForTemperature,
   stageResultsArtifacts,
   syncResultsRepo
-} from "./chunk-36HXBYUY.js";
+} from "./chunk-27VT2KU2.js";
 import {
   CLI_PLACEHOLDERS,
   COMMON_TARGET_SETTINGS,
@@ -47,6 +47,7 @@ import {
   interpolateEnv,
   isGraderKind,
   listTargetNames,
+  loadCasesFromDirectory,
   loadCasesFromFile,
   loadConfig,
   loadTestSuite,
@@ -61,12 +62,12 @@ import {
   subscribeToCopilotSdkLogEntries,
   subscribeToPiLogEntries,
   toCamelCaseDeep
-} from "./chunk-LP4Y5D2Z.js";
+} from "./chunk-ERSBQAGK.js";
 
 // package.json
 var package_default = {
   name: "agentv",
-  version: "4.20.0",
+  version: "4.21.0",
   description: "CLI entry point for AgentV",
   type: "module",
   repository: {
@@ -291,9 +292,108 @@ import path15 from "node:path";
 import { pathToFileURL } from "node:url";
 
 // src/version-check.ts
-import { coerce, satisfies, validRange } from "semver";
+import { coerce, major, satisfies, validRange } from "semver";
+
+// src/self-update.ts
+import { spawn } from "node:child_process";
+import { get } from "node:https";
+var NPM_REGISTRY_URL = "https://registry.npmjs.org/agentv/latest";
+function detectPackageManagerFromPath(scriptPath) {
+  if (scriptPath.includes(".bun")) {
+    return "bun";
+  }
+  return "npm";
+}
+function detectPackageManager() {
+  return detectPackageManagerFromPath(process.argv[1] ?? "");
+}
+function detectInstallScopeFromPath(scriptPath) {
+  const hasSegment = scriptPath.includes("/node_modules/") || scriptPath.includes("\\node_modules\\");
+  return hasSegment ? "local" : "global";
+}
+function detectInstallScope() {
+  return detectInstallScopeFromPath(process.argv[1] ?? "");
+}
+function runCommand(cmd, args) {
+  return new Promise((resolve, reject) => {
+    const child = spawn(cmd, args, { stdio: ["inherit", "pipe", "inherit"] });
+    let stdout = "";
+    child.stdout?.on("data", (data) => {
+      process.stdout.write(data);
+      stdout += data.toString();
+    });
+    child.on("error", reject);
+    child.on("close", (code) => resolve({ exitCode: code ?? 1, stdout }));
+  });
+}
+function fetchLatestVersion() {
+  return new Promise((resolve) => {
+    const req = get(NPM_REGISTRY_URL, { timeout: 5e3 }, (res) => {
+      if (res.statusCode !== 200) {
+        res.resume();
+        resolve(null);
+        return;
+      }
+      let body = "";
+      res.on("data", (chunk) => {
+        body += chunk.toString();
+      });
+      res.on("end", () => {
+        try {
+          const version = JSON.parse(body).version;
+          resolve(typeof version === "string" ? version : null);
+        } catch {
+          resolve(null);
+        }
+      });
+    });
+    req.on("error", () => resolve(null));
+    req.on("timeout", () => {
+      req.destroy();
+      resolve(null);
+    });
+  });
+}
+function getInstallArgs(pm, versionSpec, scope) {
+  const pkg = `agentv@${versionSpec}`;
+  const baseCmd = pm === "npm" ? "install" : "add";
+  return scope === "global" ? [baseCmd, "-g", pkg] : [baseCmd, pkg];
+}
+async function performSelfUpdate(options) {
+  const pm = options?.pm ?? detectPackageManager();
+  const currentVersion = options?.currentVersion ?? "unknown";
+  const versionSpec = options?.versionRange ?? "latest";
+  const scope = options?.scope ?? detectInstallScope();
+  const args = getInstallArgs(pm, versionSpec, scope);
+  try {
+    const result = await runCommand(pm, args);
+    if (result.exitCode !== 0) {
+      return { success: false, currentVersion, scope };
+    }
+    let newVersion;
+    try {
+      const versionResult = await runCommand("agentv", ["--version"]);
+      newVersion = versionResult.stdout.trim();
+    } catch {
+    }
+    return { success: true, currentVersion, newVersion, scope };
+  } catch (error) {
+    if (error instanceof Error) {
+      if (error.message.includes("ENOENT") || error.message.includes("not found")) {
+        const alternative = pm === "npm" ? "bun" : "npm";
+        console.error(`Error: ${pm} not found. Try using --${alternative} flag.`);
+      } else {
+        console.error(`Error: ${error.message}`);
+      }
+    }
+    return { success: false, currentVersion, scope };
+  }
+}
+
+// src/version-check.ts
 var ANSI_YELLOW = "\x1B[33m";
 var ANSI_RED = "\x1B[31m";
+var ANSI_GREEN = "\x1B[32m";
 var ANSI_RESET = "\x1B[0m";
 function checkVersion(requiredVersion) {
   const currentVersion = package_default.version;
@@ -319,10 +419,10 @@ async function enforceRequiredVersion(requiredVersion, options) {
   if (result.satisfied) {
     return;
   }
-  const warning = `${ANSI_YELLOW}Warning: This project requires agentv ${result.requiredRange} but you have ${result.currentVersion}.${ANSI_RESET}
-  Run \`agentv self update\` to upgrade.`;
+  const warning = `${ANSI_YELLOW}Warning: This project requires agentv ${result.requiredRange} but you have ${result.currentVersion}.${ANSI_RESET}`;
   if (options?.strict) {
-    console.error(warning);
+    console.error(`${warning}
+  Run \`agentv self update\` to upgrade.`);
     console.error(
       `${ANSI_RED}Aborting: --strict mode requires the installed version to satisfy the required range.${ANSI_RESET}`
     );
@@ -330,18 +430,40 @@ async function enforceRequiredVersion(requiredVersion, options) {
   }
   if (process.stdin.isTTY && process.stdout.isTTY) {
     console.warn(warning);
-    const shouldContinue = await promptContinue();
-    if (!shouldContinue) {
-      process.exit(1);
+    const shouldUpdate = await promptUpdate();
+    if (shouldUpdate) {
+      await runInlineUpdate(result.currentVersion, result.requiredRange);
     }
   } else {
     process.stderr.write(`${warning}
+  Run \`agentv self update\` to upgrade.
 `);
   }
 }
-async function promptContinue() {
+async function promptUpdate() {
   const { confirm } = await import("@inquirer/prompts");
-  return confirm({ message: "Continue anyway?", default: false });
+  return confirm({ message: "Update now?", default: true });
+}
+async function runInlineUpdate(currentVersion, versionRange) {
+  const currentMajor = major(coerce(currentVersion) ?? currentVersion);
+  const safeRange = `${versionRange} <${currentMajor + 1}.0.0`;
+  console.log("");
+  const result = await performSelfUpdate({ currentVersion, versionRange: safeRange });
+  if (!result.success) {
+    console.error(`${ANSI_RED}Update failed. Run \`agentv self update\` manually.${ANSI_RESET}`);
+    process.exit(1);
+  }
+  if (result.newVersion) {
+    console.log(
+      `
+${ANSI_GREEN}Update complete: ${currentVersion} \u2192 ${result.newVersion}${ANSI_RESET}`
+    );
+  } else {
+    console.log(`
+${ANSI_GREEN}Update complete.${ANSI_RESET}`);
+  }
+  console.log("Please re-run your command.");
+  process.exit(0);
 }
 
 // src/commands/results/remote.ts
@@ -2212,7 +2334,7 @@ function createWriterFromPath(filePath, options) {
 
 // src/commands/eval/progress-display.ts
 var ANSI_BOLD = "\x1B[1m";
-var ANSI_GREEN = "\x1B[32m";
+var ANSI_GREEN2 = "\x1B[32m";
 var ANSI_RED2 = "\x1B[31m";
 var ANSI_YELLOW2 = "\x1B[33m";
 var ANSI_RESET2 = "\x1B[0m";
@@ -2226,7 +2348,7 @@ function formatVerdict(score, verdict) {
   const scoreStr = score !== void 0 ? `${Math.round(score * 100)}%` : "";
   const verdictLabel = verdict === "ERROR" ? "ERROR" : `${scoreStr} ${verdict}`;
   if (!colors2) return ` | ${verdictLabel}`;
-  const color = verdict === "PASS" ? ANSI_GREEN : verdict === "FAIL" ? ANSI_RED2 : ANSI_YELLOW2;
+  const color = verdict === "PASS" ? ANSI_GREEN2 : verdict === "FAIL" ? ANSI_RED2 : ANSI_YELLOW2;
   return ` | ${color}${ANSI_BOLD}${verdictLabel}${ANSI_RESET2}`;
 }
 var ProgressDisplay = class {
@@ -2644,7 +2766,7 @@ function formatMatrixSummary(results) {
 import { readFile as readFile2 } from "node:fs/promises";
 import path14 from "node:path";
 import { parse } from "yaml";
-import { readFile as readFile22, readdir } from "node:fs/promises";
+import { readFile as readFile22, readdir, stat as stat2 } from "node:fs/promises";
 import path22 from "node:path";
 import { parse as parse2 } from "yaml";
 import { readFile as readFile3 } from "node:fs/promises";
@@ -2880,17 +3002,21 @@ async function validateEvalFile(filePath) {
   }
   const cases = parsed.tests;
   if (typeof cases === "string") {
-    validateTestsStringPath(cases, absolutePath, errors);
     await validateWorkspaceConfig(parsed.workspace, absolutePath, errors, "workspace");
-    const ext = path22.extname(cases).toLowerCase();
-    if (VALID_TEST_FILE_EXTENSIONS.has(ext)) {
-      const externalCasesPath = path22.resolve(path22.dirname(absolutePath), cases);
+    const externalCasesPath = path22.resolve(path22.dirname(absolutePath), cases);
+    let isDir = false;
+    try {
+      const pathStat = await stat2(externalCasesPath);
+      isDir = pathStat.isDirectory();
+    } catch {
+    }
+    if (isDir) {
       try {
-        const externalCases = await loadCasesFromFile(externalCasesPath);
-        for (let i = 0; i < externalCases.length; i++) {
-          const externalCase = externalCases[i];
+        const dirCases = await loadCasesFromDirectory(externalCasesPath);
+        for (let i = 0; i < dirCases.length; i++) {
+          const dirCase = dirCases[i];
           await validateWorkspaceConfig(
-            externalCase.workspace,
+            dirCase.workspace,
             absolutePath,
             errors,
             `tests[${i}].workspace`
@@ -2905,6 +3031,31 @@ async function validateEvalFile(filePath) {
           message
         });
       }
+    } else {
+      validateTestsStringPath(cases, absolutePath, errors);
+      const ext = path22.extname(cases).toLowerCase();
+      if (VALID_TEST_FILE_EXTENSIONS.has(ext)) {
+        try {
+          const externalCases = await loadCasesFromFile(externalCasesPath);
+          for (let i = 0; i < externalCases.length; i++) {
+            const externalCase = externalCases[i];
+            await validateWorkspaceConfig(
+              externalCase.workspace,
+              absolutePath,
+              errors,
+              `tests[${i}].workspace`
+            );
+          }
+        } catch (error) {
+          const message = error instanceof Error ? error.message : String(error);
+          errors.push({
+            severity: "error",
+            filePath: absolutePath,
+            location: "tests",
+            message
+          });
+        }
+      }
     }
     return {
       valid: errors.filter((e) => e.severity === "error").length === 0,
@@ -5224,8 +5375,8 @@ async function runEvalCommand(input) {
   if (options.workspacePath) {
     const resolvedWorkspace = path15.resolve(options.workspacePath);
     try {
-      const { stat: stat2 } = await import("node:fs/promises");
-      const stats = await stat2(resolvedWorkspace);
+      const { stat: stat3 } = await import("node:fs/promises");
+      const stats = await stat3(resolvedWorkspace);
       if (!stats.isDirectory()) {
         throw new Error(`--workspace-path is not a directory: ${resolvedWorkspace}`);
       }
@@ -5281,7 +5432,7 @@ async function runEvalCommand(input) {
   const useFileExport = !!options.otelFile;
   if (options.exportOtel || useFileExport) {
     try {
-      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-GURCO6IS.js");
+      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-RSAA3T6F.js");
       let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
       let headers = {};
       if (options.otelBackend) {
@@ -5480,7 +5631,7 @@ async function runEvalCommand(input) {
   const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
   let transcriptProviderFactory;
   if (options.transcript) {
-    const { TranscriptProvider } = await import("./dist-GURCO6IS.js");
+    const { TranscriptProvider } = await import("./dist-RSAA3T6F.js");
     const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
     const totalTests = [...fileMetadata.values()].reduce(
       (sum, meta) => sum + meta.testCases.length,
@@ -5649,7 +5800,7 @@ async function runEvalCommand(input) {
     if (usesDefaultArtifactWorkspace && allResults.length > 0) {
       const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
       if (isResumeAppend) {
-        const { writePerTestArtifacts } = await import("./artifact-writer-RFXWXUOV.js");
+        const { writePerTestArtifacts } = await import("./artifact-writer-E775664W.js");
         await writePerTestArtifacts(allResults, runDir, {
           experiment: normalizeExperimentName(options.experiment)
         });
@@ -5830,6 +5981,11 @@ export {
   loadManifestResults,
   loadLightweightResults,
   HtmlWriter,
+  detectPackageManager,
+  detectInstallScope,
+  fetchLatestVersion,
+  performSelfUpdate,
+  enforceRequiredVersion,
   resolveEvalPaths,
   findRepoRoot,
   c,
@@ -5867,4 +6023,4 @@ export {
   getCategories,
   filterByCategory
 };
-//# sourceMappingURL=chunk-PHGEGHKR.js.map
+//# sourceMappingURL=chunk-GPGX24OI.js.map