@agentv/core 4.27.0-next.1 → 4.28.0-next.1

package/dist/index.d.cts

@@ -1796,6 +1796,8 @@ interface ClaudeResolvedConfig {
     readonly logFormat?: 'summary' | 'json';
     /** New stream_log field. false=no stream log (default), 'raw'=per-event, 'summary'=consolidated. */
     readonly streamLog?: false | 'raw' | 'summary';
+    /** When true (default), passes --dangerously-skip-permissions to the Claude CLI. Matches ClaudeSdkProvider behavior. */
+    readonly bypassPermissions?: boolean;
 }
 interface MockResolvedConfig {
     readonly response?: string;
@@ -4269,9 +4271,16 @@ declare function getWorkspacePoolRoot(): string;
  *     - id: my-app
  *       name: My App
  *       path: /home/user/projects/my-app
+ *       source:
+ *         url: ${{ BENCHMARK_REPO_URL }}
+ *         ref: ${{ BENCHMARK_REPO_REF:-main }}
  *       added_at: "2026-03-20T10:00:00Z"
  *       last_opened_at: "2026-03-30T14:00:00Z"
  *
+ * The optional `source` field enables remote sync via syncBenchmarks():
+ *   first run — git clone --depth 1 --filter=blob:none
+ *   subsequent runs — git pull --ff-only
+ *
  * Concurrency: the registry assumes a single writer. All mutating calls
  * (add/remove/touchBenchmark) do read-modify-write on benchmarks.yaml
  * without a lock. Studio's HTTP handlers are serialized by Node's
@@ -4284,12 +4293,17 @@ declare function getWorkspacePoolRoot(): string;
  *   - discoverBenchmarks() is a one-shot filesystem utility for bulk
  *     registration; it does not run in the request path.
  */
+interface BenchmarkSource {
+    url: string;
+    ref: string;
+}
 interface BenchmarkEntry {
     id: string;
     name: string;
     path: string;
     addedAt: string;
     lastOpenedAt: string;
+    source?: BenchmarkSource;
 }
 interface BenchmarkRegistry {
     benchmarks: BenchmarkEntry[];
@@ -4328,6 +4342,34 @@ declare function touchBenchmark(benchmarkId: string): void;
  */
 declare function discoverBenchmarks(rootDir: string, maxDepth?: number): string[];
 
+/**
+ * Benchmark sync — pulls remote git repos to the local path declared in the
+ * benchmark registry before Studio/eval startup.
+ *
+ * Sync is oneshot only, triggered by the Studio UI "Sync" button or the
+ * `agentv benchmark sync` CLI command. There is no daemon or continuous mode.
+ *
+ *   First run  — git clone --depth 1 --filter=blob:none --branch <ref> <url> <path>
+ *   Subsequent — git pull --ff-only (when <path>/.git already exists)
+ *
+ * Usage:
+ *   import { syncBenchmarks } from './benchmark-sync.js';
+ *   await syncBenchmarks(registry.benchmarks);
+ */
+
+/**
+ * Clone or pull a single benchmark entry from its declared source.
+ * - No .git present: shallow clone into entry.path.
+ * - .git present: git pull --ff-only to update in place.
+ * Throws on git error or missing source.
+ */
+declare function syncBenchmark(entry: BenchmarkEntry): Promise<void>;
+/**
+ * Iterate benchmark entries and sync any that have a source declared.
+ * Entries without source are skipped silently.
+ */
+declare function syncBenchmarks(entries: BenchmarkEntry[]): Promise<void>;
+
 /**
  * Trims an EvaluationResult for baseline storage.
  * Strips large debug/audit fields (denylist approach) while preserving
@@ -4900,4 +4942,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceEnvConfig, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directPushResults, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, killAllTrackedChildren, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseEnvOutput, parseJsonFromText, parseJsonSafe, parseYamlValue, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runBeforeSessionHook, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreRangeEvaluationSchema, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, trackChild, trackedChildCount, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };
+export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, type BenchmarkSource, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceEnvConfig, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directPushResults, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, killAllTrackedChildren, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseEnvOutput, parseJsonFromText, parseJsonSafe, parseYamlValue, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runBeforeSessionHook, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreRangeEvaluationSchema, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncBenchmark, syncBenchmarks, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, trackChild, trackedChildCount, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };

package/dist/index.d.ts

@@ -1796,6 +1796,8 @@ interface ClaudeResolvedConfig {
     readonly logFormat?: 'summary' | 'json';
     /** New stream_log field. false=no stream log (default), 'raw'=per-event, 'summary'=consolidated. */
     readonly streamLog?: false | 'raw' | 'summary';
+    /** When true (default), passes --dangerously-skip-permissions to the Claude CLI. Matches ClaudeSdkProvider behavior. */
+    readonly bypassPermissions?: boolean;
 }
 interface MockResolvedConfig {
     readonly response?: string;
@@ -4269,9 +4271,16 @@ declare function getWorkspacePoolRoot(): string;
  *     - id: my-app
  *       name: My App
  *       path: /home/user/projects/my-app
+ *       source:
+ *         url: ${{ BENCHMARK_REPO_URL }}
+ *         ref: ${{ BENCHMARK_REPO_REF:-main }}
  *       added_at: "2026-03-20T10:00:00Z"
  *       last_opened_at: "2026-03-30T14:00:00Z"
  *
+ * The optional `source` field enables remote sync via syncBenchmarks():
+ *   first run — git clone --depth 1 --filter=blob:none
+ *   subsequent runs — git pull --ff-only
+ *
  * Concurrency: the registry assumes a single writer. All mutating calls
  * (add/remove/touchBenchmark) do read-modify-write on benchmarks.yaml
  * without a lock. Studio's HTTP handlers are serialized by Node's
@@ -4284,12 +4293,17 @@ declare function getWorkspacePoolRoot(): string;
  *   - discoverBenchmarks() is a one-shot filesystem utility for bulk
  *     registration; it does not run in the request path.
  */
+interface BenchmarkSource {
+    url: string;
+    ref: string;
+}
 interface BenchmarkEntry {
     id: string;
     name: string;
     path: string;
     addedAt: string;
     lastOpenedAt: string;
+    source?: BenchmarkSource;
 }
 interface BenchmarkRegistry {
     benchmarks: BenchmarkEntry[];
@@ -4328,6 +4342,34 @@ declare function touchBenchmark(benchmarkId: string): void;
  */
 declare function discoverBenchmarks(rootDir: string, maxDepth?: number): string[];
 
+/**
+ * Benchmark sync — pulls remote git repos to the local path declared in the
+ * benchmark registry before Studio/eval startup.
+ *
+ * Sync is oneshot only, triggered by the Studio UI "Sync" button or the
+ * `agentv benchmark sync` CLI command. There is no daemon or continuous mode.
+ *
+ *   First run  — git clone --depth 1 --filter=blob:none --branch <ref> <url> <path>
+ *   Subsequent — git pull --ff-only (when <path>/.git already exists)
+ *
+ * Usage:
+ *   import { syncBenchmarks } from './benchmark-sync.js';
+ *   await syncBenchmarks(registry.benchmarks);
+ */
+
+/**
+ * Clone or pull a single benchmark entry from its declared source.
+ * - No .git present: shallow clone into entry.path.
+ * - .git present: git pull --ff-only to update in place.
+ * Throws on git error or missing source.
+ */
+declare function syncBenchmark(entry: BenchmarkEntry): Promise<void>;
+/**
+ * Iterate benchmark entries and sync any that have a source declared.
+ * Entries without source are skipped silently.
+ */
+declare function syncBenchmarks(entries: BenchmarkEntry[]): Promise<void>;
+
 /**
  * Trims an EvaluationResult for baseline storage.
  * Strips large debug/audit fields (denylist approach) while preserving
@@ -4900,4 +4942,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceEnvConfig, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directPushResults, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, killAllTrackedChildren, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseEnvOutput, parseJsonFromText, parseJsonSafe, parseYamlValue, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runBeforeSessionHook, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreRangeEvaluationSchema, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, trackChild, trackedChildCount, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };
+export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, type BenchmarkSource, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceEnvConfig, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directPushResults, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, killAllTrackedChildren, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseEnvOutput, parseJsonFromText, parseJsonSafe, parseYamlValue, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runBeforeSessionHook, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreRangeEvaluationSchema, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncBenchmark, syncBenchmarks, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, trackChild, trackedChildCount, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };

package/dist/index.js

@@ -132,7 +132,7 @@ import {
   tokensPerTool,
   trackChild,
   trackedChildCount
-} from "./chunk-JNBHD34F.js";
+} from "./chunk-SCC35F3L.js";
 import {
   COMMON_TARGET_SETTINGS,
   TEST_MESSAGE_ROLES,
@@ -157,9 +157,9 @@ import {
   resolveDelegatedTargetDefinition,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-7LQI7772.js";
+} from "./chunk-YFXMMBUG.js";
 import "./chunk-3WGHC7LC.js";
-import "./chunk-M5X2KMEA.js";
+import "./chunk-YDFZ7XN3.js";
 import {
   OtlpJsonFileExporter
 } from "./chunk-KPSI5CSL.js";
@@ -486,12 +486,12 @@ var CONFIG_FILE_NAMES = [
   ".agentv/config.js"
 ];
 async function loadTsConfig(projectRoot) {
-  const { existsSync: existsSync3 } = await import("node:fs");
+  const { existsSync: existsSync4 } = await import("node:fs");
   const { pathToFileURL } = await import("node:url");
   const { join } = await import("node:path");
   for (const fileName of CONFIG_FILE_NAMES) {
     const filePath = join(projectRoot, fileName);
-    if (!existsSync3(filePath)) {
+    if (!existsSync4(filePath)) {
       continue;
     }
     try {
@@ -1053,22 +1053,33 @@ function fromYaml(raw) {
   if (typeof e.id !== "string" || typeof e.name !== "string" || typeof e.path !== "string") {
     return null;
   }
-  return {
+  const entry = {
     id: e.id,
     name: e.name,
     path: e.path,
     addedAt: typeof e.added_at === "string" ? e.added_at : "",
     lastOpenedAt: typeof e.last_opened_at === "string" ? e.last_opened_at : ""
   };
+  if (e.source && typeof e.source === "object") {
+    const s = e.source;
+    if (typeof s.url === "string" && typeof s.ref === "string") {
+      entry.source = { url: s.url, ref: s.ref };
+    }
+  }
+  return entry;
 }
 function toYaml(entry) {
-  return {
+  const yaml = {
     id: entry.id,
     name: entry.name,
     path: entry.path,
     added_at: entry.addedAt,
     last_opened_at: entry.lastOpenedAt
   };
+  if (entry.source) {
+    yaml.source = { url: entry.source.url, ref: entry.source.ref };
+  }
+  return yaml;
 }
 function loadBenchmarkRegistry() {
   const registryPath = getBenchmarksRegistryPath();
@@ -1081,7 +1092,8 @@ function loadBenchmarkRegistry() {
     if (!parsed || typeof parsed !== "object") {
       return { benchmarks: [] };
     }
-    const benchmarks = Array.isArray(parsed.benchmarks) ? parsed.benchmarks.map(fromYaml).filter((e) => e !== null) : [];
+    const env = process.env;
+    const benchmarks = Array.isArray(parsed.benchmarks) ? parsed.benchmarks.map((e) => fromYaml(interpolateEnv(e, env))).filter((e) => e !== null) : [];
     return { benchmarks };
   } catch {
     return { benchmarks: [] };
@@ -1180,6 +1192,34 @@ function discoverBenchmarks(rootDir, maxDepth = 2) {
   return results.sort();
 }
 
+// src/benchmark-sync.ts
+import * as childProcess from "node:child_process";
+import { existsSync as existsSync3 } from "node:fs";
+async function syncBenchmark(entry) {
+  if (!entry.source) {
+    throw new Error(`Benchmark '${entry.id}' has no source defined`);
+  }
+  const { url, ref } = entry.source;
+  const dest = entry.path;
+  if (existsSync3(`${dest}/.git`)) {
+    childProcess.execFileSync("git", ["-C", dest, "pull", "--ff-only"], { stdio: "inherit" });
+  } else {
+    childProcess.execFileSync(
+      "git",
+      ["clone", "--depth", "1", "--filter=blob:none", "--branch", ref, url, dest],
+      { stdio: "inherit" }
+    );
+  }
+}
+async function syncBenchmarks(entries) {
+  for (const entry of entries) {
+    if (!entry.source) continue;
+    console.log(`Syncing benchmark '${entry.id}' from ${entry.source.url}...`);
+    await syncBenchmark(entry);
+    console.log(`Benchmark '${entry.id}' synced.`);
+  }
+}
+
 // src/evaluation/baseline.ts
 var STRIPPED_TOP_LEVEL_FIELDS = /* @__PURE__ */ new Set([
   "requests",
@@ -2629,6 +2669,8 @@ export {
   subscribeToCopilotSdkLogEntries,
   subscribeToPiLogEntries,
   substituteVariables,
+  syncBenchmark,
+  syncBenchmarks,
   syncResultsRepo,
   toCamelCaseDeep,
   toSnakeCaseDeep,