@agentv/core 4.21.0-next.1 → 4.22.0-next.1

package/dist/index.cjs

@@ -788,11 +788,13 @@ async function loadConfig(evalFilePath, repoRoot) {
         configPath
       );
       const results = parseResultsConfig(parsed.results, configPath);
+      const hooks = parseHooksConfig(parsed.hooks, configPath);
       return {
         required_version: requiredVersion,
         eval_patterns: evalPatterns,
         execution: executionDefaults,
-        results
+        results,
+        ...hooks && { hooks }
       };
     } catch (error) {
       logWarning(
@@ -1139,6 +1141,25 @@ function parseResultsExportConfig(raw, configPath) {
     ...branchPrefix && { branch_prefix: branchPrefix }
   };
 }
+function parseHooksConfig(raw, configPath) {
+  if (raw === void 0 || raw === null) {
+    return void 0;
+  }
+  if (typeof raw !== "object" || Array.isArray(raw)) {
+    logWarning(`Invalid hooks in ${configPath}, expected object`);
+    return void 0;
+  }
+  const obj = raw;
+  const beforeSession = obj.before_session;
+  if (beforeSession !== void 0) {
+    if (typeof beforeSession !== "string" || beforeSession.trim().length === 0) {
+      logWarning(`Invalid hooks.before_session in ${configPath}, expected non-empty string`);
+      return void 0;
+    }
+    return { before_session: beforeSession.trim() };
+  }
+  return void 0;
+}
 function logWarning(message) {
   console.warn(`${ANSI_YELLOW2}Warning: ${message}${ANSI_RESET3}`);
 }
@@ -24440,6 +24461,7 @@ __export(index_exports, {
   parseClaudeSession: () => parseClaudeSession,
   parseCodexSession: () => parseCodexSession,
   parseCopilotEvents: () => parseCopilotEvents,
+  parseEnvOutput: () => parseEnvOutput,
   parseJsonFromText: () => parseJsonFromText,
   parseJsonSafe: () => parseJsonSafe,
   prepareResultsRepoBranch: () => prepareResultsRepoBranch,
@@ -24459,6 +24481,7 @@ __export(index_exports, {
   resolveTargetDefinition: () => resolveTargetDefinition,
   resolveWorkspaceTemplate: () => resolveWorkspaceTemplate,
   rubricEvaluationSchema: () => rubricEvaluationSchema,
+  runBeforeSessionHook: () => runBeforeSessionHook,
   runContainsAllAssertion: () => runContainsAllAssertion,
   runContainsAnyAssertion: () => runContainsAnyAssertion,
   runContainsAssertion: () => runContainsAssertion,
@@ -24474,6 +24497,7 @@ __export(index_exports, {
   runStartsWithAssertion: () => runStartsWithAssertion,
   saveBenchmarkRegistry: () => saveBenchmarkRegistry,
   scanRepoDeps: () => scanRepoDeps,
+  scoreRangeEvaluationSchema: () => scoreRangeEvaluationSchema,
   scoreToVerdict: () => scoreToVerdict,
   shouldEnableCache: () => shouldEnableCache,
   shouldSkipCacheForTemperature: () => shouldSkipCacheForTemperature,
@@ -24811,6 +24835,17 @@ var AgentVConfigSchema = import_zod5.z.object({
     maxCostUsd: import_zod5.z.number().min(0).optional(),
     /** Maximum duration per run in milliseconds */
     maxDurationMs: import_zod5.z.number().int().min(0).optional()
+  }).optional(),
+  /** Lifecycle hooks */
+  hooks: import_zod5.z.object({
+    /**
+     * Shell command to run once at agentv startup, before any command executes.
+     * stdout is parsed for env var exports (`KEY=value` or `export KEY="value"`)
+     * and injected into process.env. Keys already set in the environment are
+     * not overwritten — existing env always takes priority.
+     * stderr is forwarded to the user. Non-zero exit aborts with an error.
+     */
+    beforeSession: import_zod5.z.string().optional()
   }).optional()
 });
 function defineConfig(config) {
@@ -26110,6 +26145,63 @@ var RunBudgetTracker = class {
   }
 };
 
+// src/evaluation/hooks.ts
+init_cjs_shims();
+var import_node_child_process13 = require("child_process");
+var ANSI_YELLOW9 = "\x1B[33m";
+var ANSI_RESET10 = "\x1B[0m";
+function parseEnvOutput(stdout) {
+  const result = {};
+  for (const line of stdout.split("\n")) {
+    const trimmed = line.trim();
+    if (!trimmed || trimmed.startsWith("#")) continue;
+    const match = trimmed.match(/^(?:export\s+)?([A-Za-z_][A-Za-z0-9_]*)=(.*)$/);
+    if (!match) continue;
+    const key = match[1];
+    let value = match[2];
+    if (value.startsWith('"') && value.endsWith('"') || value.startsWith("'") && value.endsWith("'")) {
+      value = value.slice(1, -1);
+    }
+    if (key) {
+      result[key] = value;
+    }
+  }
+  return result;
+}
+function runBeforeSessionHook(command) {
+  const isWindows = process.platform === "win32";
+  const shell = isWindows ? "cmd" : "sh";
+  const shellFlag = isWindows ? "/c" : "-c";
+  console.log(`${ANSI_YELLOW9}Running before_session hook: ${command}${ANSI_RESET10}`);
+  const result = (0, import_node_child_process13.spawnSync)(shell, [shellFlag, command], {
+    encoding: "utf8",
+    // Do not inherit stdio — capture stdout for parsing, forward stderr manually
+    stdio: ["ignore", "pipe", "pipe"]
+  });
+  if (result.stderr) {
+    process.stderr.write(result.stderr);
+  }
+  if (result.error) {
+    throw new Error(`before_session hook failed to start: ${result.error.message}`);
+  }
+  if (result.status !== 0) {
+    throw new Error(
+      `before_session hook exited with code ${result.status ?? "unknown"}: ${command}`
+    );
+  }
+  const vars = parseEnvOutput(result.stdout ?? "");
+  let injected = 0;
+  for (const [key, value] of Object.entries(vars)) {
+    if (process.env[key] === void 0) {
+      process.env[key] = value;
+      injected++;
+    }
+  }
+  if (injected > 0) {
+    console.log(`before_session hook injected ${injected} environment variable(s).`);
+  }
+}
+
 // src/import/index.ts
 init_cjs_shims();
 
@@ -26896,6 +26988,7 @@ function createAgentKernel() {
   parseClaudeSession,
   parseCodexSession,
   parseCopilotEvents,
+  parseEnvOutput,
   parseJsonFromText,
   parseJsonSafe,
   prepareResultsRepoBranch,
@@ -26915,6 +27008,7 @@ function createAgentKernel() {
   resolveTargetDefinition,
   resolveWorkspaceTemplate,
   rubricEvaluationSchema,
+  runBeforeSessionHook,
   runContainsAllAssertion,
   runContainsAnyAssertion,
   runContainsAssertion,
@@ -26930,6 +27024,7 @@ function createAgentKernel() {
   runStartsWithAssertion,
   saveBenchmarkRegistry,
   scanRepoDeps,
+  scoreRangeEvaluationSchema,
   scoreToVerdict,
   shouldEnableCache,
   shouldSkipCacheForTemperature,