fix(core): unwrap YAML merge keys (<<:) in eval loader

apps/cli/src/commands/results/studio-config.ts

@@ -21,8 +21,8 @@
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
 import path from 'node:path';
 
-import { DEFAULT_THRESHOLD } from '@agentv/core';
-import { parse as parseYaml, stringify as stringifyYaml } from 'yaml';
+import { DEFAULT_THRESHOLD, parseYamlValue } from '@agentv/core';
+import { stringify as stringifyYaml } from 'yaml';
 
 export interface StudioConfig {
   threshold: number;
@@ -47,7 +47,7 @@ export function loadStudioConfig(agentvDir: string): StudioConfig {
   }
 
   const raw = readFileSync(configPath, 'utf-8');
-  const parsed = parseYaml(raw);
+  const parsed = parseYamlValue(raw);
 
   if (!parsed || typeof parsed !== 'object') {
     return { ...DEFAULTS };
@@ -89,7 +89,7 @@ export function saveStudioConfig(agentvDir: string, config: StudioConfig): void
   let existing: Record<string, unknown> = {};
   if (existsSync(configPath)) {
     const raw = readFileSync(configPath, 'utf-8');
-    const parsed = parseYaml(raw);
+    const parsed = parseYamlValue(raw);
     if (parsed && typeof parsed === 'object' && !Array.isArray(parsed)) {
       existing = parsed as Record<string, unknown>;
     }

packages/core/src/benchmarks.ts

@@ -32,8 +32,9 @@
 import { existsSync, mkdirSync, readFileSync, readdirSync, statSync, writeFileSync } from 'node:fs';
 import path from 'node:path';
 
-import { parse as parseYaml, stringify as stringifyYaml } from 'yaml';
+import { stringify as stringifyYaml } from 'yaml';
 
+import { parseYamlValue } from './evaluation/yaml-loader.js';
 import { getAgentvConfigDir } from './paths.js';
 
 // ── Types ───────────────────────────────────────────────────────────────
@@ -101,7 +102,7 @@ export function loadBenchmarkRegistry(): BenchmarkRegistry {
   }
   try {
     const raw = readFileSync(registryPath, 'utf-8');
-    const parsed = parseYaml(raw);
+    const parsed = parseYamlValue(raw) as { benchmarks?: unknown } | null | undefined;
     if (!parsed || typeof parsed !== 'object') {
       return { benchmarks: [] };
     }

packages/core/src/evaluation/loaders/case-file-loader.ts

@@ -1,11 +1,11 @@
 import { readFile, readdir, stat } from 'node:fs/promises';
 import path from 'node:path';
 import fg from 'fast-glob';
-import { parse as parseYaml } from 'yaml';
 
 import { interpolateEnv } from '../interpolation.js';
 import type { JsonObject, JsonValue } from '../types.js';
 import { isJsonObject } from '../types.js';
+import { parseYamlValue } from '../yaml-loader.js';
 
 const ANSI_YELLOW = '\u001b[33m';
 const ANSI_RESET = '\u001b[0m';
@@ -38,7 +38,7 @@ function isGlobPattern(filePath: string): boolean {
  * Expects the file to contain an array of test objects.
  */
 function parseYamlCases(content: string, filePath: string): JsonObject[] {
-  const raw = parseYaml(content) as unknown;
+  const raw = parseYamlValue(content);
   const parsed = interpolateEnv(raw, process.env);
   if (!Array.isArray(parsed)) {
     throw new Error(
@@ -206,7 +206,7 @@ export async function loadCasesFromDirectory(dirPath: string): Promise<JsonObjec
       throw new Error(`Cannot read case file: ${caseFilePath}\n  ${message}`);
     }
 
-    const raw = parseYaml(content) as unknown;
+    const raw = parseYamlValue(content);
     const parsed = interpolateEnv(raw, process.env);
     if (!isJsonObject(parsed)) {
       throw new Error(

packages/core/src/evaluation/loaders/config-loader.ts

@@ -1,8 +1,8 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
 import { interpolateEnv } from '../interpolation.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import type {
   EvalTargetRef,
   FailOnError,
@@ -77,7 +77,7 @@ export async function loadConfig(
 
     try {
       const rawConfig = await readFile(configPath, 'utf8');
-      const parsed = interpolateEnv(parse(rawConfig), process.env) as unknown;
+      const parsed = interpolateEnv(parseYamlValue(rawConfig), process.env) as unknown;
 
       if (!isJsonObject(parsed)) {
         logWarning(`Invalid .agentv/config.yaml format at ${configPath}`);

packages/core/src/evaluation/loaders/eval-yaml-transpiler.ts

@@ -9,7 +9,8 @@
 
 import { readFileSync } from 'node:fs';
 import path from 'node:path';
-import { parse } from 'yaml';
+
+import { parseYamlValue } from '../yaml-loader.js';
 
 // ---------------------------------------------------------------------------
 // evals.json output types
@@ -510,7 +511,7 @@ export function transpileEvalYaml(suite: unknown, source = 'EVAL.yaml'): Transpi
  */
 export function transpileEvalYamlFile(evalYamlPath: string): TranspileResult {
   const content = readFileSync(evalYamlPath, 'utf8');
-  const parsed = parse(content) as unknown;
+  const parsed = parseYamlValue(content);
   return transpileEvalYaml(parsed, path.basename(evalYamlPath));
 }
 

packages/core/src/evaluation/loaders/grader-parser.ts

@@ -1,9 +1,9 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
 import { normalizePreprocessorType } from '../content-preprocessor.js';
 import { interpolateEnv } from '../interpolation.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ToolTrajectoryExpectedItem, ToolTrajectoryGraderConfig } from '../trace.js';
 import type {
   ContentPreprocessorConfig,
@@ -183,7 +183,7 @@ async function loadAssertionTemplateEntries(
   }
 
   const content = await readFile(resolved.resolvedPath, 'utf8');
-  const parsed = interpolateEnv(parse(content), process.env) as unknown;
+  const parsed = interpolateEnv(parseYamlValue(content), process.env) as unknown;
   if (!isJsonObject(parsed)) {
     throw new Error(
       `Invalid assertion template file in '${evalId}': ${resolved.resolvedPath} (expected a YAML object with an assertions array)`,

packages/core/src/evaluation/loaders/jsonl-parser.ts

@@ -1,12 +1,12 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
 import micromatch from 'micromatch';
-import { parse as parseYaml } from 'yaml';
 
 import { collectResolvedInputFilePaths } from '../input-message-utils.js';
 import { interpolateEnv } from '../interpolation.js';
 import type { EvalTest, JsonObject, JsonValue, TestMessage } from '../types.js';
 import { isJsonObject, isTestMessage } from '../types.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import { buildSearchRoots, fileExists, resolveToAbsolutePath } from './file-resolver.js';
 import {
   coerceEvaluator,
@@ -92,7 +92,7 @@ async function loadSidecarMetadata(jsonlPath: string, verbose: boolean): Promise
 
   try {
     const content = await readFile(sidecarPath, 'utf8');
-    const parsed = interpolateEnv(parseYaml(content), process.env) as unknown;
+    const parsed = interpolateEnv(parseYamlValue(content), process.env) as unknown;
 
     if (!isJsonObject(parsed)) {
       logWarning(`Invalid sidecar metadata format in ${sidecarPath}`);

packages/core/src/evaluation/providers/copilot-session-discovery.ts

@@ -16,7 +16,8 @@
 import { readFile, readdir, stat } from 'node:fs/promises';
 import { homedir } from 'node:os';
 import path from 'node:path';
-import { parse as parseYaml } from 'yaml';
+
+import { parseYamlValue } from '../yaml-loader.js';
 
 export interface CopilotSession {
   readonly sessionId: string;
@@ -60,7 +61,7 @@ export async function discoverCopilotSessions(opts?: DiscoverOptions): Promise<C
 
     try {
       const workspaceContent = await readFile(workspacePath, 'utf8');
-      const workspace = (parseYaml(workspaceContent) ?? {}) as Record<string, unknown>;
+      const workspace = (parseYamlValue(workspaceContent) ?? {}) as Record<string, unknown>;
 
       const cwd = String(workspace.cwd ?? '');
 

packages/core/src/evaluation/providers/targets-file.ts

@@ -1,8 +1,8 @@
 import { constants } from 'node:fs';
 import { access, readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
+import { parseYamlValue } from '../yaml-loader.js';
 import { TARGETS_SCHEMA_V2 } from './types.js';
 import type { TargetDefinition } from './types.js';
 
@@ -62,7 +62,7 @@ export async function readTargetDefinitions(
   }
 
   const raw = await readFile(absolutePath, 'utf8');
-  const parsed = parse(raw) as unknown;
+  const parsed = parseYamlValue(raw);
 
   if (!isRecord(parsed)) {
     throw new Error(`targets.yaml at ${absolutePath} must be a YAML object with a 'targets' field`);

packages/core/src/evaluation/validation/cases-validator.ts

@@ -1,7 +1,7 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ValidationError, ValidationResult } from './types.js';
 
 type JsonValue = string | number | boolean | null | JsonObject | JsonArray;
@@ -26,7 +26,7 @@ export async function validateCasesFile(filePath: string): Promise<ValidationRes
   let parsed: unknown;
   try {
     const content = await readFile(absolutePath, 'utf8');
-    parsed = parse(content);
+    parsed = parseYamlValue(content);
   } catch (error) {
     errors.push({
       severity: 'error',

packages/core/src/evaluation/validation/config-validator.ts

@@ -1,6 +1,6 @@
 import { readFile } from 'node:fs/promises';
-import { parse } from 'yaml';
 
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ValidationError, ValidationResult } from './types.js';
 
 /**
@@ -11,7 +11,7 @@ export async function validateConfigFile(filePath: string): Promise<ValidationRe
 
   try {
     const content = await readFile(filePath, 'utf8');
-    const parsed = parse(content) as unknown;
+    const parsed = parseYamlValue(content);
 
     // Check if parsed content is an object
     if (typeof parsed !== 'object' || parsed === null) {

packages/core/src/evaluation/validation/eval-validator.ts

@@ -1,10 +1,10 @@
 import { readFile, readdir, stat } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
 import { interpolateEnv } from '../interpolation.js';
 import { loadCasesFromDirectory, loadCasesFromFile } from '../loaders/case-file-loader.js';
 import { isGraderKind } from '../types.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ValidationError, ValidationResult } from './types.js';
 
 type JsonValue = string | number | boolean | null | JsonObject | JsonArray;
@@ -165,7 +165,7 @@ export async function validateEvalFile(filePath: string): Promise<ValidationResu
   let parsed: unknown;
   try {
     const content = await readFile(absolutePath, 'utf8');
-    parsed = interpolateEnv(parse(content), process.env);
+    parsed = interpolateEnv(parseYamlValue(content), process.env);
   } catch (error) {
     errors.push({
       severity: 'error',
@@ -514,7 +514,7 @@ async function validateWorkspaceConfig(
 
   try {
     const workspaceContent = await readFile(workspacePath, 'utf8');
-    const parsedWorkspace = interpolateEnv(parse(workspaceContent), process.env);
+    const parsedWorkspace = interpolateEnv(parseYamlValue(workspaceContent), process.env);
     if (!isObject(parsedWorkspace)) {
       errors.push({
         severity: 'error',

packages/core/src/evaluation/validation/file-reference-validator.ts

@@ -1,8 +1,8 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
 import { buildSearchRoots, findGitRoot, resolveFileReference } from '../file-utils.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ValidationError } from './types.js';
 
 type JsonValue = string | number | boolean | null | JsonObject | JsonArray;
@@ -40,7 +40,7 @@ export async function validateFileReferences(
   let parsed: unknown;
   try {
     const content = await readFile(absolutePath, 'utf8');
-    parsed = parse(content);
+    parsed = parseYamlValue(content);
   } catch {
     // Parse errors are already caught by eval-validator
     return errors;

packages/core/src/evaluation/validation/file-type.ts

@@ -1,7 +1,7 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
+import { parseYamlValue } from '../yaml-loader.js';
 import type { FileType } from './types.js';
 
 const SCHEMA_EVAL_V2 = 'agentv-eval-v2';
@@ -18,7 +18,7 @@ const SCHEMA_CONFIG_V2 = 'agentv-config-v2';
 export async function detectFileType(filePath: string): Promise<FileType> {
   try {
     const content = await readFile(filePath, 'utf8');
-    const parsed = parse(content) as unknown;
+    const parsed = parseYamlValue(content);
 
     // YAML array root → cases file (array of test case objects)
     if (Array.isArray(parsed)) {

packages/core/src/evaluation/validation/targets-validator.ts

@@ -1,13 +1,13 @@
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
 import {
   CLI_PLACEHOLDERS,
   COMMON_TARGET_SETTINGS,
   findDeprecatedCamelCaseTargetWarnings,
 } from '../providers/targets.js';
 import { KNOWN_PROVIDERS, PROVIDER_ALIASES } from '../providers/types.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ValidationError, ValidationResult } from './types.js';
 
 type JsonValue = string | number | boolean | null | JsonObject | JsonArray;
@@ -280,7 +280,7 @@ export async function validateTargetsFile(filePath: string): Promise<ValidationR
   let parsed: unknown;
   try {
     const content = await readFile(absolutePath, 'utf8');
-    parsed = parse(content);
+    parsed = parseYamlValue(content);
   } catch (error) {
     errors.push({
       severity: 'error',

packages/core/src/evaluation/validation/workspace-path-validator.ts

@@ -1,7 +1,7 @@
 import { access, readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
+import { parseYamlValue } from '../yaml-loader.js';
 import type { ValidationError } from './types.js';
 
 type JsonValue = string | number | boolean | null | JsonObject | JsonArray;
@@ -38,7 +38,7 @@ export async function validateWorkspacePaths(
   let parsed: unknown;
   try {
     const content = await readFile(absolutePath, 'utf8');
-    parsed = parse(content);
+    parsed = parseYamlValue(content);
   } catch {
     // Parse errors are already caught by eval-validator
     return errors;
@@ -55,7 +55,7 @@ export async function validateWorkspacePaths(
     const workspaceFilePath = path.resolve(evalDir, workspaceRaw);
     try {
       const wsContent = await readFile(workspaceFilePath, 'utf8');
-      const wsParsed = parse(wsContent);
+      const wsParsed = parseYamlValue(wsContent);
       if (isObject(wsParsed)) {
         const wsDir = path.dirname(workspaceFilePath);
         await validateWorkspaceObject(wsParsed, wsDir, absolutePath, 'workspace', errors);

packages/core/src/evaluation/workspace/deps-scanner.ts

@@ -17,10 +17,10 @@
  */
 import { readFile } from 'node:fs/promises';
 import path from 'node:path';
-import { parse } from 'yaml';
 
 import { interpolateEnv } from '../interpolation.js';
 import type { RepoCheckout, RepoClone, RepoSource } from '../types.js';
+import { parseYamlValue } from '../yaml-loader.js';
 import { parseRepoCheckout, parseRepoClone, parseRepoSource } from './repo-config-parser.js';
 
 /** A single git repo dependency discovered from eval files. */
@@ -115,7 +115,7 @@ interface RawRepo {
 
 async function extractReposFromEvalFile(filePath: string): Promise<RawRepo[]> {
   const content = await readFile(filePath, 'utf8');
-  const parsed = interpolateEnv(parse(content), process.env);
+  const parsed = interpolateEnv(parseYamlValue(content), process.env);
   if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) return [];
   const obj = parsed as Record<string, unknown>;
   const evalFileDir = path.dirname(path.resolve(filePath));
@@ -148,7 +148,7 @@ async function extractReposFromWorkspaceRaw(raw: unknown, evalFileDir: string):
     // External workspace file reference
     const workspaceFilePath = path.resolve(evalFileDir, raw);
     const content = await readFile(workspaceFilePath, 'utf8');
-    const parsed = interpolateEnv(parse(content), process.env);
+    const parsed = interpolateEnv(parseYamlValue(content), process.env);
     if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) return [];
     return extractReposFromObject(parsed as Record<string, unknown>);
   }