@agentv/core 4.5.1 → 4.6.0

package/dist/index.d.cts

@@ -215,7 +215,8 @@ interface Provider {
 type EnvLookup = Readonly<Record<string, string | undefined>>;
 interface TargetDefinition {
     readonly name: string;
-    readonly provider: ProviderKind | string;
+    readonly provider?: ProviderKind | string;
+    readonly use_target?: string | unknown | undefined;
     readonly grader_target?: string | undefined;
     /** @deprecated Use `grader_target` instead */
     readonly judge_target?: string | undefined;
@@ -302,6 +303,8 @@ interface TargetDefinition {
     readonly retryBackoffFactor?: number | unknown | undefined;
     readonly retry_status_codes?: unknown | undefined;
     readonly retryStatusCodes?: unknown | undefined;
+    readonly fallback_targets?: readonly string[] | unknown | undefined;
+    readonly fallbackTargets?: readonly string[] | unknown | undefined;
 }
 
 /**
@@ -1200,6 +1203,11 @@ interface EvaluationResult {
     readonly score: number;
     readonly assertions: readonly AssertionEntry[];
     readonly target: string;
+    /**
+     * The target that actually served the response, when different from the
+     * primary target. Present only when a fallback target was used.
+     */
+    readonly targetUsed?: string;
     /** Token usage metrics from provider (optional) */
     readonly tokenUsage?: TokenUsage;
     /** Total cost in USD (optional, from provider) */
@@ -1683,6 +1691,14 @@ interface RetryConfig {
     readonly backoffFactor?: number;
     readonly retryableStatusCodes?: readonly number[];
 }
+/**
+ * Selects which OpenAI-compatible API endpoint to use.
+ * - "chat" (default): POST /chat/completions — universally supported by all OpenAI-compatible providers.
+ * - "responses": POST /responses — only supported by api.openai.com.
+ *
+ * Maps to Vercel AI SDK methods: "chat" → provider.chat(model), "responses" → provider(model).
+ */
+type ApiFormat = 'chat' | 'responses';
 /**
  * Azure OpenAI settings used by the Vercel AI SDK.
  */
@@ -1702,6 +1718,7 @@ interface OpenAIResolvedConfig {
     readonly baseURL: string;
     readonly apiKey: string;
     readonly model: string;
+    readonly apiFormat?: ApiFormat;
     readonly temperature?: number;
     readonly maxOutputTokens?: number;
     readonly retry?: RetryConfig;
@@ -1852,6 +1869,11 @@ interface ResolvedTargetBase {
      * to force CLI invocation even in subagent mode.
      */
     readonly subagentModeAllowed?: boolean;
+    /**
+     * Ordered list of target names to try when the primary target fails after
+     * exhausting retries. Each fallback is attempted in order.
+     */
+    readonly fallbackTargets?: readonly string[];
 }
 type ResolvedTarget = (ResolvedTargetBase & {
     readonly kind: 'openai';
@@ -1913,7 +1935,7 @@ type ResolvedTarget = (ResolvedTargetBase & {
  * Exported so the targets validator can reuse the same list — adding a field
  * here automatically makes it valid in targets.yaml without a separate update.
  */
-declare const COMMON_TARGET_SETTINGS: readonly ["provider_batching", "providerBatching", "subagent_mode_allowed", "subagentModeAllowed"];
+declare const COMMON_TARGET_SETTINGS: readonly ["use_target", "provider_batching", "providerBatching", "subagent_mode_allowed", "subagentModeAllowed", "fallback_targets", "fallbackTargets"];
 declare function resolveTargetDefinition(definition: TargetDefinition, env?: EnvLookup, evalFilePath?: string): ResolvedTarget;
 
 /**
@@ -3887,4 +3909,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, COMMON_TARGET_SETTINGS, type CacheConfig, type ChildEvaluatorResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type Content, type ContentFile, type ContentImage, type ContentText, type CopilotCliResolvedConfig, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_CATEGORY, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type InlineAssertEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmGraderEvaluator, type LlmGraderEvaluatorConfig, type LlmGraderEvaluatorOptions, type LlmGraderPromptAssembly, LlmGraderEvaluator as LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmGraderEvaluatorOptions as LlmJudgeEvaluatorOptions, type LlmGraderPromptAssembly as LlmJudgePromptAssembly, 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 PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type ProjectEntry, type ProjectRegistry, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerEvaluator, type SkillTriggerEvaluatorConfig, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addProject, assembleLlmGraderPrompt, assembleLlmGraderPrompt as assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveCategory, deriveProjectId, detectFormat, discoverAssertions, discoverClaudeSessions, discoverCopilotSessions, discoverGraders, discoverGraders as discoverJudges, discoverProjects, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getOutputFilenames, getProject, getProjectsRegistryPath, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isEvaluatorKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadProjectRegistry, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseAgentSkillsEvals, parseClaudeSession, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, removeProject, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveProjectRegistry, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, touchProject, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };
+export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, COMMON_TARGET_SETTINGS, type CacheConfig, type ChildEvaluatorResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type Content, type ContentFile, type ContentImage, type ContentText, type CopilotCliResolvedConfig, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_CATEGORY, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type InlineAssertEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmGraderEvaluator, type LlmGraderEvaluatorConfig, type LlmGraderEvaluatorOptions, type LlmGraderPromptAssembly, LlmGraderEvaluator as LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmGraderEvaluatorOptions as LlmJudgeEvaluatorOptions, type LlmGraderPromptAssembly as LlmJudgePromptAssembly, 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 PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type ProjectEntry, type ProjectRegistry, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerEvaluator, type SkillTriggerEvaluatorConfig, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addProject, assembleLlmGraderPrompt, assembleLlmGraderPrompt as assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveCategory, deriveProjectId, detectFormat, discoverAssertions, discoverClaudeSessions, discoverCopilotSessions, discoverGraders, discoverGraders as discoverJudges, discoverProjects, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getOutputFilenames, getProject, getProjectsRegistryPath, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isEvaluatorKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadProjectRegistry, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseAgentSkillsEvals, parseClaudeSession, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, removeProject, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveProjectRegistry, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, touchProject, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };

package/dist/index.d.ts

@@ -215,7 +215,8 @@ interface Provider {
 type EnvLookup = Readonly<Record<string, string | undefined>>;
 interface TargetDefinition {
     readonly name: string;
-    readonly provider: ProviderKind | string;
+    readonly provider?: ProviderKind | string;
+    readonly use_target?: string | unknown | undefined;
     readonly grader_target?: string | undefined;
     /** @deprecated Use `grader_target` instead */
     readonly judge_target?: string | undefined;
@@ -302,6 +303,8 @@ interface TargetDefinition {
     readonly retryBackoffFactor?: number | unknown | undefined;
     readonly retry_status_codes?: unknown | undefined;
     readonly retryStatusCodes?: unknown | undefined;
+    readonly fallback_targets?: readonly string[] | unknown | undefined;
+    readonly fallbackTargets?: readonly string[] | unknown | undefined;
 }
 
 /**
@@ -1200,6 +1203,11 @@ interface EvaluationResult {
     readonly score: number;
     readonly assertions: readonly AssertionEntry[];
     readonly target: string;
+    /**
+     * The target that actually served the response, when different from the
+     * primary target. Present only when a fallback target was used.
+     */
+    readonly targetUsed?: string;
     /** Token usage metrics from provider (optional) */
     readonly tokenUsage?: TokenUsage;
     /** Total cost in USD (optional, from provider) */
@@ -1683,6 +1691,14 @@ interface RetryConfig {
     readonly backoffFactor?: number;
     readonly retryableStatusCodes?: readonly number[];
 }
+/**
+ * Selects which OpenAI-compatible API endpoint to use.
+ * - "chat" (default): POST /chat/completions — universally supported by all OpenAI-compatible providers.
+ * - "responses": POST /responses — only supported by api.openai.com.
+ *
+ * Maps to Vercel AI SDK methods: "chat" → provider.chat(model), "responses" → provider(model).
+ */
+type ApiFormat = 'chat' | 'responses';
 /**
  * Azure OpenAI settings used by the Vercel AI SDK.
  */
@@ -1702,6 +1718,7 @@ interface OpenAIResolvedConfig {
     readonly baseURL: string;
     readonly apiKey: string;
     readonly model: string;
+    readonly apiFormat?: ApiFormat;
     readonly temperature?: number;
     readonly maxOutputTokens?: number;
     readonly retry?: RetryConfig;
@@ -1852,6 +1869,11 @@ interface ResolvedTargetBase {
      * to force CLI invocation even in subagent mode.
      */
     readonly subagentModeAllowed?: boolean;
+    /**
+     * Ordered list of target names to try when the primary target fails after
+     * exhausting retries. Each fallback is attempted in order.
+     */
+    readonly fallbackTargets?: readonly string[];
 }
 type ResolvedTarget = (ResolvedTargetBase & {
     readonly kind: 'openai';
@@ -1913,7 +1935,7 @@ type ResolvedTarget = (ResolvedTargetBase & {
  * Exported so the targets validator can reuse the same list — adding a field
  * here automatically makes it valid in targets.yaml without a separate update.
  */
-declare const COMMON_TARGET_SETTINGS: readonly ["provider_batching", "providerBatching", "subagent_mode_allowed", "subagentModeAllowed"];
+declare const COMMON_TARGET_SETTINGS: readonly ["use_target", "provider_batching", "providerBatching", "subagent_mode_allowed", "subagentModeAllowed", "fallback_targets", "fallbackTargets"];
 declare function resolveTargetDefinition(definition: TargetDefinition, env?: EnvLookup, evalFilePath?: string): ResolvedTarget;
 
 /**
@@ -3887,4 +3909,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, COMMON_TARGET_SETTINGS, type CacheConfig, type ChildEvaluatorResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type Content, type ContentFile, type ContentImage, type ContentText, type CopilotCliResolvedConfig, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_CATEGORY, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type InlineAssertEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmGraderEvaluator, type LlmGraderEvaluatorConfig, type LlmGraderEvaluatorOptions, type LlmGraderPromptAssembly, LlmGraderEvaluator as LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmGraderEvaluatorOptions as LlmJudgeEvaluatorOptions, type LlmGraderPromptAssembly as LlmJudgePromptAssembly, 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 PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type ProjectEntry, type ProjectRegistry, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerEvaluator, type SkillTriggerEvaluatorConfig, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addProject, assembleLlmGraderPrompt, assembleLlmGraderPrompt as assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveCategory, deriveProjectId, detectFormat, discoverAssertions, discoverClaudeSessions, discoverCopilotSessions, discoverGraders, discoverGraders as discoverJudges, discoverProjects, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getOutputFilenames, getProject, getProjectsRegistryPath, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isEvaluatorKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadProjectRegistry, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseAgentSkillsEvals, parseClaudeSession, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, removeProject, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveProjectRegistry, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, touchProject, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };
+export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, COMMON_TARGET_SETTINGS, type CacheConfig, type ChildEvaluatorResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type Content, type ContentFile, type ContentImage, type ContentText, type CopilotCliResolvedConfig, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_CATEGORY, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type InlineAssertEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmGraderEvaluator, type LlmGraderEvaluatorConfig, type LlmGraderEvaluatorOptions, type LlmGraderPromptAssembly, LlmGraderEvaluator as LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmGraderEvaluatorOptions as LlmJudgeEvaluatorOptions, type LlmGraderPromptAssembly as LlmJudgePromptAssembly, 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 PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type ProjectEntry, type ProjectRegistry, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerEvaluator, type SkillTriggerEvaluatorConfig, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addProject, assembleLlmGraderPrompt, assembleLlmGraderPrompt as assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveCategory, deriveProjectId, detectFormat, discoverAssertions, discoverClaudeSessions, discoverCopilotSessions, discoverGraders, discoverGraders as discoverJudges, discoverProjects, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getOutputFilenames, getProject, getProjectsRegistryPath, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isEvaluatorKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadProjectRegistry, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseAgentSkillsEvals, parseClaudeSession, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, removeProject, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveProjectRegistry, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, touchProject, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };

package/dist/index.js

@@ -23,7 +23,7 @@ import {
   readTextFile,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-M65PVDQ5.js";
+} from "./chunk-AIQ5FO4G.js";
 import {
   AgentvProvider
 } from "./chunk-PRNXHNLF.js";
@@ -766,6 +766,7 @@ function validateTemplateVariables(content, source) {
 // src/evaluation/loaders/evaluator-parser.ts
 var ANSI_YELLOW3 = "\x1B[33m";
 var ANSI_RESET4 = "\x1B[0m";
+var PROMPT_FILE_PREFIX = "file://";
 function normalizeEvaluatorType(type) {
   return type.replace(/_/g, "-");
 }
@@ -1064,12 +1065,23 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
           threshold: thresholdValue
         };
       } else {
-        const aggregatorPrompt = asString(rawAggregator.prompt);
+        const rawAggPrompt = asString(rawAggregator.prompt);
+        let aggregatorPrompt;
         let promptPath2;
-        if (aggregatorPrompt) {
-          const resolved = await resolveFileReference2(aggregatorPrompt, searchRoots);
-          if (resolved.resolvedPath) {
-            promptPath2 = path4.resolve(resolved.resolvedPath);
+        if (rawAggPrompt) {
+          if (rawAggPrompt.startsWith(PROMPT_FILE_PREFIX)) {
+            const fileRef = rawAggPrompt.slice(PROMPT_FILE_PREFIX.length);
+            aggregatorPrompt = fileRef;
+            const resolved = await resolveFileReference2(fileRef, searchRoots);
+            if (resolved.resolvedPath) {
+              promptPath2 = path4.resolve(resolved.resolvedPath);
+            } else {
+              throw new Error(
+                `Composite aggregator in '${evalId}': prompt file not found: ${resolved.displayPath}`
+              );
+            }
+          } else {
+            aggregatorPrompt = rawAggPrompt;
           }
         }
         aggregator = {
@@ -1649,21 +1661,25 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         promptScriptConfig = rawPrompt.config;
       }
     } else if (typeof rawPrompt === "string") {
-      prompt = rawPrompt;
-      const resolved = await resolveFileReference2(prompt, searchRoots);
-      if (resolved.resolvedPath) {
-        promptPath = path4.resolve(resolved.resolvedPath);
-        try {
-          await validateCustomPromptContent(promptPath);
-        } catch (error) {
-          const message = error instanceof Error ? error.message : String(error);
-          throw new Error(`Evaluator '${name}' template (${promptPath}): ${message}`);
+      if (rawPrompt.startsWith(PROMPT_FILE_PREFIX)) {
+        const fileRef = rawPrompt.slice(PROMPT_FILE_PREFIX.length);
+        prompt = fileRef;
+        const resolved = await resolveFileReference2(fileRef, searchRoots);
+        if (resolved.resolvedPath) {
+          promptPath = path4.resolve(resolved.resolvedPath);
+          try {
+            await validateCustomPromptContent(promptPath);
+          } catch (error) {
+            const message = error instanceof Error ? error.message : String(error);
+            throw new Error(`Evaluator '${name}' template (${promptPath}): ${message}`);
+          }
+        } else {
+          throw new Error(
+            `Evaluator '${name}' in '${evalId}': prompt file not found: ${resolved.displayPath}`
+          );
         }
       } else {
-        logWarning2(
-          `Inline prompt used for evaluator '${name}' in '${evalId}' (file not found: ${resolved.displayPath})`,
-          resolved.attempted.length > 0 ? resolved.attempted.map((attempt) => `  Tried: ${attempt}`) : void 0
-        );
+        prompt = rawPrompt;
       }
     }
     const _model = asString(rawEvaluator.model);
@@ -3572,7 +3588,7 @@ var OpenAIProvider = class {
       apiKey: config.apiKey,
       baseURL: config.baseURL
     });
-    this.model = openai(config.model);
+    this.model = config.apiFormat === "responses" ? openai(config.model) : openai.chat(config.model);
   }
   id;
   kind = "openai";
@@ -10082,8 +10098,11 @@ function assertTargetDefinition(value, index, filePath) {
       `targets.yaml entry at index ${index} in ${filePath} is missing a valid 'name'`
     );
   }
-  if (typeof provider !== "string" || provider.trim().length === 0) {
-    throw new Error(`targets.yaml entry '${name}' in ${filePath} is missing a valid 'provider'`);
+  const hasUseTarget = typeof value.use_target === "string" && value.use_target.trim().length > 0;
+  if (!hasUseTarget && (typeof provider !== "string" || provider.trim().length === 0)) {
+    throw new Error(
+      `targets.yaml entry '${name}' in ${filePath} is missing a valid 'provider' (or use use_target for delegation)`
+    );
   }
   return value;
 }
@@ -15490,10 +15509,20 @@ async function runEvaluation(options) {
     if (resolvedTargetsByName.has(name)) {
       return resolvedTargetsByName.get(name);
     }
-    const definition = targetDefinitions.get(name);
+    let definition = targetDefinitions.get(name);
     if (!definition) {
       return void 0;
     }
+    for (let depth = 0; depth < 5; depth++) {
+      const useTarget = definition.use_target;
+      if (typeof useTarget !== "string" || useTarget.trim().length === 0) break;
+      const envMatch = useTarget.trim().match(/^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i);
+      const resolvedName = envMatch ? envLookup[envMatch[1]] ?? "" : useTarget.trim();
+      if (resolvedName.length === 0) break;
+      const next = targetDefinitions.get(resolvedName);
+      if (!next) break;
+      definition = next;
+    }
     const resolved = resolveTargetDefinition(definition, envLookup, evalFilePath);
     resolvedTargetsByName.set(name, resolved);
     return resolved;
@@ -16498,6 +16527,7 @@ async function runEvalCase(options) {
   let attempt = 0;
   let providerResponse = cachedResponse;
   let lastError;
+  let targetUsed;
   while (!providerResponse && attempt < attemptBudget) {
     try {
       providerResponse = await invokeProvider(provider, {
@@ -16520,25 +16550,33 @@ async function runEvalCase(options) {
         attempt += 1;
         continue;
       }
-      const errorResult = buildErrorResult(
-        evalCase,
-        target.name,
-        nowFn(),
-        error,
-        promptInputs,
-        provider,
-        "agent",
-        "provider_error",
-        verbose
-      );
-      if (workspacePath) {
-        if (forceCleanup) {
-          await cleanupWorkspace(workspacePath).catch(() => {
-          });
-        }
-        return { ...errorResult, workspacePath };
+      break;
+    }
+  }
+  if (!providerResponse && target.fallbackTargets?.length && targetResolver) {
+    for (const fallbackName of target.fallbackTargets) {
+      const fallbackProvider = targetResolver(fallbackName);
+      if (!fallbackProvider) {
+        continue;
+      }
+      try {
+        providerResponse = await invokeProvider(fallbackProvider, {
+          evalCase,
+          target,
+          promptInputs,
+          attempt: 0,
+          agentTimeoutMs,
+          signal,
+          cwd: workspacePath,
+          workspaceFile: caseWorkspaceFile ?? suiteWorkspaceFile,
+          captureFileChanges: !!baselineCommit,
+          streamCallbacks: options.streamCallbacks
+        });
+        targetUsed = fallbackName;
+        break;
+      } catch (error) {
+        lastError = error;
       }
-      return errorResult;
     }
   }
   if (!providerResponse) {
@@ -16664,8 +16702,10 @@ async function runEvalCase(options) {
     };
     const skippedEvaluatorError = buildSkippedEvaluatorError(result.scores);
     const executionStatus = providerError || skippedEvaluatorError ? "execution_error" : classifyQualityStatus(result.score, caseThreshold);
+    const targetUsedField = targetUsed ? { targetUsed } : {};
     const finalResult = providerError ? {
       ...result,
+      ...targetUsedField,
       evalRun,
       error: providerError,
       executionStatus,
@@ -16677,6 +16717,7 @@ async function runEvalCase(options) {
       afterEachOutput
     } : skippedEvaluatorError ? {
       ...result,
+      ...targetUsedField,
       score: 0,
       evalRun,
       error: skippedEvaluatorError,
@@ -16689,6 +16730,7 @@ async function runEvalCase(options) {
       afterEachOutput
     } : {
       ...result,
+      ...targetUsedField,
       evalRun,
       executionStatus,
       beforeAllOutput,