promptfoo 0.121.7 → 0.121.9

package/dist/src/server/index.js

@@ -1,50 +1,50 @@
 import { t as __exportAll } from "../chunk-DEq-mXcV.js";
-import { C as getEnvFloat, D as getMaxEvalTimeMs, E as getEvalTimeoutMs, O as isCI, S as getEnvBool, T as getEnvString, a as logger, b as summarizeEvaluateResultForLogging, f as sanitizeObject, g as getAjv, h as extractJsonObjects, k as state, m as extractFirstJsonObject, n as globalLogCallback, o as setLogCallback, r as isDebugEnabled, s as setLogLevel, t as getLogLevel, v as orderKeys, w as getEnvInt, y as safeJsonStringify } from "../logger-KD8JjCRJ.js";
-import { F as getDefaultShareViewBaseUrl, I as getShareApiBaseUrl, L as getShareViewBaseUrl, N as TERMINAL_MAX_WIDTH, P as getDefaultPort, R as FILE_METADATA_KEY, S as isPromptfooSampleTarget, T as parseChatPrompt, c as CloudConfig, l as cloudConfig, m as sleep, n as fetchWithRetries, p as getCurrentTimestamp, r as fetchWithTimeout, t as fetchWithProxy, v as REQUEST_TIMEOUT_MS, z as HUMAN_ASSERTION_TYPE } from "../fetch-BufrQtvR.js";
-import { n as VERSION } from "../version-BK20a4sw.js";
+import { C as getEnvBool, D as getEvalTimeoutMs, E as getEnvString, O as getMaxEvalTimeMs, S as state, T as getEnvInt, a as logger, b as summarizeEvaluateResultForLogging, f as sanitizeObject, g as getAjv, h as extractJsonObjects, k as isCI, m as extractFirstJsonObject, n as globalLogCallback, o as setLogCallback, r as isDebugEnabled, s as setLogLevel, t as getLogLevel, v as orderKeys, w as getEnvFloat, y as safeJsonStringify } from "../logger-BjYLHvGm.js";
+import { F as getDefaultShareViewBaseUrl, I as getShareApiBaseUrl, L as getShareViewBaseUrl, N as TERMINAL_MAX_WIDTH, P as getDefaultPort, R as FILE_METADATA_KEY, S as isPromptfooSampleTarget, T as parseChatPrompt, c as CloudConfig, l as cloudConfig, m as sleep, n as fetchWithRetries, p as getCurrentTimestamp, r as fetchWithTimeout, t as fetchWithProxy, y as getRequestTimeoutMs, z as HUMAN_ASSERTION_TYPE } from "../fetch-CBe8pU8Q.js";
+import { n as VERSION } from "../version-BGiEZvX0.js";
 import { t as invariant } from "../invariant-DIYf9sP1.js";
-import { a as getAuthor, c as isLoggedIntoCloud, l as promptForEmailUnverified, n as checkEmailStatusAndMaybeExit, o as getUserEmail, r as clearUserEmail, s as getUserId, t as checkEmailStatus, u as setUserEmail } from "../accounts-DanM1wq_.js";
-import { a as safeJoin, r as importModule, t as getDirectory } from "../esm-B6whoAcf.js";
-import { a as extractVariablesFromTemplates, i as extractVariablesFromTemplate, n as renderVarsInObject, o as getNunjucksEngine, r as analyzeTemplateReference, t as renderEnvOnlyInObject } from "../render-_6ur1fhE.js";
-import { t as providerRegistry } from "../providerRegistry-1gB5vtzQ.js";
-import { a as getRemoteHealthUrl, l as shouldGenerateRemote, n as getRemoteGenerationExplicitlyDisabledError, r as getRemoteGenerationUrl, s as neverGenerateRemote } from "../remoteGeneration-COpWcmWd.js";
-import { a as openBrowser, i as checkServerRunning, n as BrowserBehaviorNames, s as promptYesNo, t as BrowserBehavior } from "../server-Ddp8GNMp.js";
-import { a as evalResultsTable, c as evalsToPromptsTable, d as promptsTable, f as spansTable, g as getDbSignalPath, h as getDb, i as datasetsTable, l as evalsToTagsTable, m as tracesTable, n as blobReferencesTable, o as evalsTable, p as tagsTable, r as configsTable, s as evalsToDatasetsTable, t as blobAssetsTable, u as modelAuditsTable } from "../tables-CxjU7bBd.js";
-import { $ as MULTI_INPUT_EXCLUDED_PLUGINS, A as STRATEGY_COLLECTIONS, B as ALIASED_PLUGIN_MAPPINGS, E as ALL_STRATEGIES, F as isMultiTurnStrategy, G as DEFAULT_PLUGINS, H as BIAS_PLUGINS, I as Severity, J as HARM_PLUGINS, K as FINANCIAL_PLUGINS, L as categoryAliases, M as getDefaultNFanout, N as isCustomStrategy, O as DEFAULT_STRATEGIES, P as isFanoutStrategy, Q as MEDICAL_PLUGINS, S as StrategyConfigSchema, T as AGENTIC_STRATEGIES, U as CANARY_BREAKING_STRATEGY_IDS, V as ALL_PLUGINS, W as DATASET_EXEMPT_PLUGINS, X as LLAMA_GUARD_ENABLED_CATEGORIES, Y as INSURANCE_PLUGINS, Z as LLAMA_GUARD_REPLICATE_PROVIDER, _ as ProvidersSchema, a as EvaluateOptionsSchema, at as REDTEAM_PROVIDER_HARM_PLUGINS, b as PluginConfigSchema, c as TestSuiteConfigSchema, ct as TEEN_SAFETY_PLUGINS, d as isGradingResult, dt as CODING_AGENT_CORE_PLUGINS, et as MULTI_INPUT_VAR, f as isResultFailureReason, ft as CODING_AGENT_PLUGINS, g as ProviderOptionsSchema, h as RedteamConfigSchema, ht as PromptSchema, i as EvalResultsFilterMode, it as REDTEAM_MODEL, j as STRATEGY_COLLECTION_MAPPINGS, k as MULTI_TURN_STRATEGIES, l as TestSuiteSchema, lt as TELECOM_PLUGINS, m as isProviderOptions, n as BaseAssertionTypesSchema, nt as PII_PLUGINS, ot as REMOTE_ONLY_PLUGIN_IDS, p as isApiProvider, q as FOUNDATION_PLUGINS, r as CommandLineOptionsSchema, rt as PLUGIN_CATEGORIES, s as ResultFailureReason, st as STRATEGY_EXEMPT_PLUGINS, t as AssertionOrSetSchema, tt as PHARMACY_PLUGINS, u as UnifiedConfigSchema, ut as UNALIGNED_PROVIDER_HARM_PLUGINS, v as ConversationMessageSchema, w as isUuid, y as PartialGenerationError, z as riskCategorySeverityMap } from "../types-BJQBBPTP.js";
+import { c as isLoggedIntoCloud, i as getAuthor, l as promptForEmailUnverified, n as checkEmailStatusAndMaybeExit, o as getUserEmail, r as clearUserEmail, s as getUserId, t as checkEmailStatus, u as setUserEmail } from "../accounts-BAajbKSh.js";
+import { a as safeJoin, r as importModule, t as getDirectory } from "../esm-CGER3lko.js";
+import { a as extractVariablesFromTemplates, i as extractVariablesFromTemplate, n as renderVarsInObject, o as getNunjucksEngine, r as analyzeTemplateReference, t as renderEnvOnlyInObject } from "../render-BoxrvB84.js";
+import { t as providerRegistry } from "../providerRegistry-v8Wa3f9z.js";
+import { a as getRemoteHealthUrl, l as shouldGenerateRemote, n as getRemoteGenerationExplicitlyDisabledError, r as getRemoteGenerationUrl, s as neverGenerateRemote } from "../remoteGeneration-CoxtxLvB.js";
+import { a as openBrowser, i as checkServerRunning, n as BrowserBehaviorNames, s as promptYesNo, t as BrowserBehavior } from "../server-8in_Vrrf.js";
+import { a as evalResultsTable, c as evalsToPromptsTable, d as promptsTable, f as spansTable, g as getDbSignalPath, h as getDb, i as datasetsTable, l as evalsToTagsTable, m as tracesTable, n as blobReferencesTable, o as evalsTable, p as tagsTable, r as configsTable, s as evalsToDatasetsTable, t as blobAssetsTable, u as modelAuditsTable } from "../tables-PWOqfbI_.js";
+import { $ as MULTI_INPUT_EXCLUDED_PLUGINS, A as STRATEGY_COLLECTIONS, B as ALIASED_PLUGIN_MAPPINGS, E as ALL_STRATEGIES, F as isMultiTurnStrategy, G as DEFAULT_PLUGINS, H as BIAS_PLUGINS, I as Severity, J as HARM_PLUGINS, K as FINANCIAL_PLUGINS, L as categoryAliases, M as getDefaultNFanout, N as isCustomStrategy, O as DEFAULT_STRATEGIES, P as isFanoutStrategy, Q as MEDICAL_PLUGINS, S as StrategyConfigSchema, T as AGENTIC_STRATEGIES, U as CANARY_BREAKING_STRATEGY_IDS, V as ALL_PLUGINS, W as DATASET_EXEMPT_PLUGINS, X as LLAMA_GUARD_ENABLED_CATEGORIES, Y as INSURANCE_PLUGINS, Z as LLAMA_GUARD_REPLICATE_PROVIDER, _ as ProvidersSchema, a as EvaluateOptionsSchema, at as REDTEAM_PROVIDER_HARM_PLUGINS, b as PluginConfigSchema, c as TestSuiteConfigSchema, ct as TEEN_SAFETY_PLUGINS, d as isGradingResult, dt as CODING_AGENT_CORE_PLUGINS, et as MULTI_INPUT_VAR, f as isResultFailureReason, ft as CODING_AGENT_PLUGINS, g as ProviderOptionsSchema, h as RedteamConfigSchema, ht as PromptSchema, i as EvalResultsFilterMode, it as REDTEAM_MODEL, j as STRATEGY_COLLECTION_MAPPINGS, k as MULTI_TURN_STRATEGIES, l as TestSuiteSchema, lt as TELECOM_PLUGINS, m as isProviderOptions, n as BaseAssertionTypesSchema, nt as PII_PLUGINS, ot as REMOTE_ONLY_PLUGIN_IDS, p as isApiProvider, q as FOUNDATION_PLUGINS, r as CommandLineOptionsSchema, rt as PLUGIN_CATEGORIES, s as ResultFailureReason, st as STRATEGY_EXEMPT_PLUGINS, t as AssertionOrSetSchema, tt as PHARMACY_PLUGINS, u as UnifiedConfigSchema, ut as UNALIGNED_PROVIDER_HARM_PLUGINS, v as ConversationMessageSchema, w as isUuid, y as PartialGenerationError, z as riskCategorySeverityMap } from "../types-uLCKnRiK.js";
 import { a as isImageFile, i as isAudioFile, o as isJavascriptFile, s as isVideoFile } from "../shared-BoG7qLMv.js";
 import { n as sha256, t as randomSequence } from "../createHash-CGVzWdjj.js";
-import { a as generateIdFromPrompt, t as hashPrompt } from "../utils-B0lzitHZ.js";
-import { c as NON_TRANSIENT_HTTP_STATUSES, i as getCache, l as isNonTransientHttpStatus, n as disableCache, r as fetchWithCache, s as withCacheNamespace, t as cache_exports } from "../cache-roFAE0cI.js";
+import { a as generateIdFromPrompt, t as hashPrompt } from "../utils-DMWnIu_V.js";
+import { c as NON_TRANSIENT_HTTP_STATUSES, i as getCache, l as isNonTransientHttpStatus, n as disableCache, r as fetchWithCache, s as withCacheNamespace, t as cache_exports } from "../cache-x9llZxbx.js";
 import { a as createEmptyTokenUsage, i as createEmptyAssertions, n as accumulateResponseTokenUsage, o as normalizeTokenUsage, r as accumulateTokenUsage, t as accumulateAssertionTokenUsage } from "../tokenUsageUtils-BjVkdk18.js";
-import { n as getBlobUrl, t as getBlobByHash } from "../blobs-CeFdPn_T.js";
-import { n as isBlobStorageEnabled, t as extractAndStoreBinaryData } from "../extractor-BR7XAzAL.js";
-import { a as storeMedia, c as getEvalConfigFromCloud, h as resolveTeamId, i as retrieveMedia, l as getOrgContext, m as makeRequest$1, n as isMediaStorageEnabled, o as checkCloudPermissions, p as isCloudProvider, r as mediaExists, s as getCloudDatabaseId, t as getMediaStorage, u as getPluginSeverityOverridesFromCloud } from "../storage-QdU-SmvD.js";
-import { n as telemetry, t as TelemetryEventSchema } from "../telemetry-DmXYcJNV.js";
-import { r as runPython } from "../pythonUtils-CLCgQ9tt.js";
-import { A as maybeLoadConfigFromExternalFile, C as isProviderAllowed, E as normalizeProviderRef, I as readFilters, L as readOutput, P as maybeLoadToolsFromExternalFile, R as loadFunction, S as isOpenAiProvider, _ as checkProviderApiKeys, a as evalTableToJson, b as isAnthropicProvider, c as getEvalTablePromptStrippedPayload, d as setupEnv, f as deduplicateTestCases, g as resultIsForTestCase, h as getTestCaseDeduplicationKey, i as ComparisonEvalNotFoundError, j as maybeLoadFromExternalFile, l as mergeComparisonTables, m as filterRuntimeVars, n as writeMultipleOutputs, o as generateEvalCsv, p as extractRuntimeVars, r as writeOutput, s as getEvalTableOutputPromptLocationsBySize, t as printBorder, u as fetchCsvFromGoogleSheet, v as doesProviderRefMatch, x as isGoogleProvider, y as getProviderDescription, z as parseFileUrl } from "../util-I-Rf-KaD.js";
-import { t as OpenAiChatCompletionProvider } from "../chat-CUCorGiL.js";
-import { h as validateFunctionCall } from "../transform-BIMynQsA.js";
-import { $ as SUGGEST_PROMPTS_SYSTEM_MESSAGE, A as ExcessiveAgencyPlugin, At as DEFAULT_ANTHROPIC_MODEL, B as isGraderFailure, C as PlinyPlugin, Ct as getGradingProvider, D as ImitationPlugin, Dt as retryWithDeduplication, E as IntentPlugin, Et as getCustomPolicies, F as BeavertailsPlugin, G as matchesPiScore, H as matchesFactuality, I as AegisPlugin, J as processPrompts, K as matchesTrajectoryGoalSuccess, L as RedteamGraderBase, M as DebugAccessPlugin, N as CrossSessionLeakPlugin, O as HarmbenchPlugin, Ot as sampleArray, P as ContractPlugin, Q as SELECT_BEST_PROMPT, R as RedteamPluginBase, S as makeInlinePolicyIdSync, St as getAndCheckProvider, T as OverreliancePlugin, Tt as withProviderCallExecutionContext, U as matchesGEval, V as matchesClosedQa, W as matchesLlmRubric, X as readProviderPromptMap, Y as readPrompts, Z as DEFAULT_WEB_SEARCH_PROMPT, _ as PromptExtractionPlugin, _t as coerceString, a as VLGuardPlugin, at as CONTEXT_RECALL_NOT_ATTRIBUTED_TOKEN, b as determinePolicyTypeFromId, bt as processFileReference, c as ToxicChatPlugin, ct as loadRubricPrompt, d as TeenSafetyDangerousRoleplayPlugin, dt as dotProduct, et as ANSWER_RELEVANCY_GENERATE, f as TeenSafetyDangerousContentPlugin, ft as euclideanDistance, g as RbacPlugin, gt as tryParse, h as ShellInjectionPlugin, ht as splitIntoSentences, i as VLSUPlugin, it as CONTEXT_RECALL_ATTRIBUTED_TOKEN, j as DivergentRepetitionPlugin, k as HallucinationPlugin, kt as getDefaultProviders, l as ToolDiscoveryPlugin, lt as renderLlmRubricPrompt, m as SqlInjectionPlugin, mt as normalizeMatcherTokenUsage, n as getGraderById, nt as CONTEXT_FAITHFULNESS_NLI_STATEMENTS, o as UnverifiableClaimsPlugin, ot as CONTEXT_RELEVANCE, p as TeenSafetyAgeRestrictedGoodsAndServicesPlugin, pt as fail, q as doRemoteGrading, rt as CONTEXT_RECALL, s as UnsafeBenchPlugin, st as CONTEXT_RELEVANCE_BAD, t as GRADERS, tt as CONTEXT_FAITHFULNESS_LONGFORM, u as TeenSafetyHarmfulBodyIdealsPlugin, ut as cosineSimilarity, v as PoliticsPlugin, vt as getFinalTest, w as getPiiLeakTestsForCategory, wt as getProviderCallExecutionContext, x as isValidPolicyObject, xt as callProviderWithContext, y as PolicyPlugin, yt as loadFromJavaScriptFile, z as fetchHuggingFaceDataset } from "../graders-C0nXU_ZP.js";
-import { a as resolveProvider, b as AIStudioChatProvider, d as createTransformRequest, f as createTransformResponse, l as MCPProvider, n as loadApiProvider, o as resolveProviderConfigs, p as GoogleLiveProvider, r as loadApiProviders, t as getProviderIds, u as HttpProvider, y as VertexChatProvider } from "../providers-BCCz6_IX.js";
-import { l as validateFunctionCall$1 } from "../util-efByNxcr.js";
+import { n as getBlobUrl, t as getBlobByHash } from "../blobs-ikh75MtF.js";
+import { n as isBlobStorageEnabled, t as extractAndStoreBinaryData } from "../extractor-DAnauX2V.js";
+import { a as storeMedia, c as getEvalConfigFromCloud, h as resolveTeamId, i as retrieveMedia, l as getOrgContext, m as makeRequest$1, n as isMediaStorageEnabled, o as checkCloudPermissions, p as isCloudProvider, r as mediaExists, s as getCloudDatabaseId, t as getMediaStorage, u as getPluginSeverityOverridesFromCloud } from "../storage-BC9TzmHh.js";
+import { n as telemetry, t as TelemetryEventSchema } from "../telemetry-D3yQCekN.js";
+import { r as runPython } from "../pythonUtils-CBe7VV6f.js";
+import { A as maybeLoadConfigFromExternalFile, C as isProviderAllowed, E as normalizeProviderRef, I as readFilters, L as readOutput, P as maybeLoadToolsFromExternalFile, R as loadFunction, S as isOpenAiProvider, _ as checkProviderApiKeys, a as evalTableToJson, b as isAnthropicProvider, c as getEvalTablePromptStrippedPayload, d as setupEnv, f as deduplicateTestCases, g as resultIsForTestCase, h as getTestCaseDeduplicationKey, i as ComparisonEvalNotFoundError, j as maybeLoadFromExternalFile, l as mergeComparisonTables, m as filterRuntimeVars, n as writeMultipleOutputs, o as generateEvalCsv, p as extractRuntimeVars, r as writeOutput, s as getEvalTableOutputPromptLocationsBySize, t as printBorder, u as fetchCsvFromGoogleSheet, v as doesProviderRefMatch, x as isGoogleProvider, y as getProviderDescription, z as parseFileUrl } from "../util-BGfk7zh7.js";
+import { t as OpenAiChatCompletionProvider } from "../chat-CvQMD-mp.js";
+import { h as validateFunctionCall } from "../transform-D7QlmswG.js";
+import { $ as SUGGEST_PROMPTS_SYSTEM_MESSAGE, A as ExcessiveAgencyPlugin, At as DEFAULT_ANTHROPIC_MODEL, B as isGraderFailure, C as PlinyPlugin, Ct as getGradingProvider, D as ImitationPlugin, Dt as retryWithDeduplication, E as IntentPlugin, Et as getCustomPolicies, F as BeavertailsPlugin, G as matchesPiScore, H as matchesFactuality, I as AegisPlugin, J as processPrompts, K as matchesTrajectoryGoalSuccess, L as RedteamGraderBase, M as DebugAccessPlugin, N as CrossSessionLeakPlugin, O as HarmbenchPlugin, Ot as sampleArray, P as ContractPlugin, Q as SELECT_BEST_PROMPT, R as RedteamPluginBase, S as makeInlinePolicyIdSync, St as getAndCheckProvider, T as OverreliancePlugin, Tt as withProviderCallExecutionContext, U as matchesGEval, V as matchesClosedQa, W as matchesLlmRubric, X as readProviderPromptMap, Y as readPrompts, Z as DEFAULT_WEB_SEARCH_PROMPT, _ as PromptExtractionPlugin, _t as coerceString, a as VLGuardPlugin, at as CONTEXT_RECALL_NOT_ATTRIBUTED_TOKEN, b as determinePolicyTypeFromId, bt as processFileReference, c as ToxicChatPlugin, ct as loadRubricPrompt, d as TeenSafetyDangerousRoleplayPlugin, dt as dotProduct, et as ANSWER_RELEVANCY_GENERATE, f as TeenSafetyDangerousContentPlugin, ft as euclideanDistance, g as RbacPlugin, gt as tryParse, h as ShellInjectionPlugin, ht as splitIntoSentences, i as VLSUPlugin, it as CONTEXT_RECALL_ATTRIBUTED_TOKEN, j as DivergentRepetitionPlugin, k as HallucinationPlugin, kt as getDefaultProviders, l as ToolDiscoveryPlugin, lt as renderLlmRubricPrompt, m as SqlInjectionPlugin, mt as normalizeMatcherTokenUsage, n as getGraderById, nt as CONTEXT_FAITHFULNESS_NLI_STATEMENTS, o as UnverifiableClaimsPlugin, ot as CONTEXT_RELEVANCE, p as TeenSafetyAgeRestrictedGoodsAndServicesPlugin, pt as fail, q as doRemoteGrading, rt as CONTEXT_RECALL, s as UnsafeBenchPlugin, st as CONTEXT_RELEVANCE_BAD, t as GRADERS, tt as CONTEXT_FAITHFULNESS_LONGFORM, u as TeenSafetyHarmfulBodyIdealsPlugin, ut as cosineSimilarity, v as PoliticsPlugin, vt as getFinalTest, w as getPiiLeakTestsForCategory, wt as getProviderCallExecutionContext, x as isValidPolicyObject, xt as callProviderWithContext, y as PolicyPlugin, yt as loadFromJavaScriptFile, z as fetchHuggingFaceDataset } from "../graders-k8qmhrPu.js";
+import { a as resolveProvider, b as AIStudioChatProvider, d as createTransformRequest, f as createTransformResponse, l as MCPProvider, n as loadApiProvider, o as resolveProviderConfigs, p as GoogleLiveProvider, r as loadApiProviders, t as getProviderIds, u as HttpProvider, y as VertexChatProvider } from "../providers-DDLQc0tl.js";
+import { l as validateFunctionCall$1 } from "../util-C6d3eV0A.js";
 import { t as ellipsize } from "../text-CZr46tp_.js";
 import { t as getProcessShim } from "../processShim-C_z3aRvF.js";
-import { n as loadFromPackage, t as isPackagePath } from "../packageParser-hwwSGnAZ.js";
-import { n as runRuby } from "../rubyUtils-DdGojpfv.js";
-import { n as createPlaceholderInputValue, r as materializeInputVariablesWithMetadata, t as buildPromptInputDescriptions } from "../inputVariables-B0qUChbV.js";
-import { a as extractPromptFromTags, c as isBasicRefusal, i as extractMaterializedVariablesFromJsonWithMetadata, n as extractGoalFromPrompt, o as getSessionId, r as extractInputVarsFromPrompt, s as getShortPluginId } from "../util-Df8YMvS1.js";
-import { n as PromptfooHarmfulCompletionProvider } from "../promptfoo-BDrfT30-.js";
-import { f as redteamProviderManager, g as createProviderRateLimitOptions, h as createRateLimitRegistry, m as TokenUsageTracker } from "../shared-7pmVZLNO.js";
-import { a as getTransformLabel, i as getTransformErrorMessage, n as TRANSFORM_KEYS, o as transform, r as TransformInputType, t as INLINE_FUNCTION_LABEL } from "../transform-CrPGTsij.js";
-import { i as throwIfTargetPromptExceedsMaxChars, n as getGeneratedPromptOverLimit, r as getMaxCharsPerMessageModifierValue, t as MAX_CHARS_PER_MESSAGE_MODIFIER_KEY } from "../promptLength-0qIHyhA5.js";
-import { n as checkExfilTracking, t as addIndirectWebPwnTestCases } from "../indirectWebPwn-tNx9OZ35.js";
+import { n as loadFromPackage, t as isPackagePath } from "../packageParser-D7YWfcSe.js";
+import { n as runRuby } from "../rubyUtils-DFzPvVD4.js";
+import { n as createPlaceholderInputValue, t as buildPromptInputDescriptions } from "../inputVariables-DpOsOnzU.js";
+import { a as extractPromptFromTags, c as isBasicRefusal, i as extractMaterializedVariablesFromJsonWithMetadata, n as extractGoalFromPrompt, o as getSessionId, s as getShortPluginId } from "../util-CKZReNsz.js";
+import { a as assertRemoteMaterializationHandled, l as requiresRemoteMaterialization, n as PromptfooHarmfulCompletionProvider } from "../promptfoo-DVKqidd8.js";
+import { f as redteamProviderManager, g as createProviderRateLimitOptions, h as createRateLimitRegistry, m as TokenUsageTracker } from "../shared-yBQATkD5.js";
+import { a as getTransformLabel, i as getTransformErrorMessage, n as TRANSFORM_KEYS, o as transform, r as TransformInputType, t as INLINE_FUNCTION_LABEL } from "../transform-C5rpIuDz.js";
+import { i as throwIfTargetPromptExceedsMaxChars, n as getGeneratedPromptOverLimit, r as getMaxCharsPerMessageModifierValue, t as MAX_CHARS_PER_MESSAGE_MODIFIER_KEY } from "../promptLength-5B4_P93z.js";
+import { n as checkExfilTracking, t as addIndirectWebPwnTestCases } from "../indirectWebPwn-DRYpRTx6.js";
 import dotenv from "dotenv";
-import { AsyncResource } from "node:async_hooks";
 import * as fs$3 from "fs";
 import fs, { createWriteStream, existsSync, readFileSync } from "fs";
 import * as path$2 from "path";
 import path, { join, parse } from "path";
 import chalk from "chalk";
+import { AsyncResource } from "node:async_hooks";
 import * as os$1 from "os";
 import os, { homedir } from "os";
 import yaml from "js-yaml";
@@ -4691,7 +4691,7 @@ async function startOtlpReceiverIfNeeded(testSuite) {
 		telemetry.record("feature_used", { feature: "tracing" });
 		try {
 			logger.debug("[EvaluatorTracing] Tracing configuration detected, starting OTLP receiver");
-			const { startOTLPReceiver } = await import("../otlpReceiver-B2eaKC8C.js");
+			const { startOTLPReceiver } = await import("../otlpReceiver-BkB7tsTT.js");
 			const port = testSuite.tracing.otlp.http.port || 4318;
 			const host = testSuite.tracing.otlp.http.host || "127.0.0.1";
 			const acceptFormats = normalizeOtlpAcceptFormats(testSuite.tracing.otlp.http.acceptFormats);
@@ -4715,7 +4715,7 @@ async function startOtlpReceiverIfNeeded(testSuite) {
 async function stopOtlpReceiverIfNeeded() {
 	if (otlpReceiverStarted) try {
 		logger.debug("[EvaluatorTracing] Stopping OTLP receiver");
-		const { stopOTLPReceiver } = await import("../otlpReceiver-B2eaKC8C.js");
+		const { stopOTLPReceiver } = await import("../otlpReceiver-BkB7tsTT.js");
 		await stopOTLPReceiver();
 		otlpReceiverStarted = false;
 		logger.info("[EvaluatorTracing] OTLP receiver stopped successfully");
@@ -7045,7 +7045,7 @@ async function loadWebSearchProvider(preferAnthropic = false) {
 	};
 	const loadOpenAIWebSearch = async () => {
 		try {
-			return await loadApiProvider("openai:responses:gpt-5.4-2026-03-05", { options: { config: { tools: [{ type: "web_search_preview" }] } } });
+			return await loadApiProvider("openai:responses:gpt-5.5-2026-04-23", { options: { config: { tools: [{ type: "web_search_preview" }] } } });
 		} catch (err) {
 			logger.debug(`Failed to load OpenAI web search provider: ${err}`);
 			return null;
@@ -8396,7 +8396,7 @@ async function runAssertion({ prompt, provider, assertion, test, vars, latencyMs
 			};
 		}
 		else if (filePath.endsWith(".rb")) try {
-			const { runRuby } = await import("../rubyUtils-DdGojpfv.js").then((n) => n.t);
+			const { runRuby } = await import("../rubyUtils-DFzPvVD4.js").then((n) => n.t);
 			valueFromScript = await runRuby(filePath, functionName || "get_assert", [output, context]);
 			logger.debug(`Ruby script ${filePath} output: ${valueFromScript}`);
 		} catch (error) {
@@ -10722,7 +10722,7 @@ async function resolveDefaultTestProvider(defaultTest, testCase) {
 	const defaultProvider = defaultTest.provider;
 	if (isApiProvider(defaultProvider)) return defaultProvider;
 	if (typeof defaultProvider === "object" && defaultProvider.id) {
-		const { loadApiProvider } = await import("../providers-BCCz6_IX.js").then((n) => n.i);
+		const { loadApiProvider } = await import("../providers-DDLQc0tl.js").then((n) => n.i);
 		return loadApiProvider(typeof defaultProvider.id === "function" ? defaultProvider.id() : defaultProvider.id, { options: defaultProvider });
 	}
 	return defaultProvider;
@@ -13813,7 +13813,7 @@ async function fetchRemoteGeneration(task, prompts) {
 			method: "POST",
 			headers: { "Content-Type": "application/json" },
 			body: JSON.stringify(body)
-		}, REQUEST_TIMEOUT_MS, "json");
+		}, getRequestTimeoutMs(), "json");
 		return RedTeamGenerationResponse.parse(response.data).result;
 	} catch (error) {
 		logger.warn(`Error using remote generation for task '${task}': ${error}`);
@@ -13986,7 +13986,7 @@ const DATASET_URL$2 = `https://raw.githubusercontent.com/promptfoo/promptfoo/ref
 const DATASET_URL_MULTILINGUAL = `https://raw.githubusercontent.com/promptfoo/promptfoo/refs/tags/${DATASET_VERSION}/examples/cyberseceval/prompt_injection_multilingual.json`;
 async function fetchDataset$2(limit, isMultilingual) {
 	try {
-		const response = await fetchWithTimeout(isMultilingual ? DATASET_URL_MULTILINGUAL : DATASET_URL$2, {}, REQUEST_TIMEOUT_MS);
+		const response = await fetchWithTimeout(isMultilingual ? DATASET_URL_MULTILINGUAL : DATASET_URL$2, {}, getRequestTimeoutMs());
 		if (!response.ok) throw new Error(`[CyberSecEval] HTTP status: ${response.status} ${response.statusText}`);
 		const data = await response.json();
 		if (!data || !Array.isArray(data)) throw new Error(`[CyberSecEval] Invalid response from ${DATASET_URL$2}`);
@@ -14054,7 +14054,7 @@ async function fetchDataset$1(limit) {
 		logger.debug(`[DoNotAnswer] Fetching dataset from ${DATASET_URL$1}`);
 		let csvData;
 		if (DATASET_URL$1.startsWith("http")) {
-			const response = await fetchWithTimeout(DATASET_URL$1, {}, REQUEST_TIMEOUT_MS);
+			const response = await fetchWithTimeout(DATASET_URL$1, {}, getRequestTimeoutMs());
 			if (!response.ok) throw new Error(`[DoNotAnswer] HTTP status: ${response.status} ${response.statusText}`);
 			csvData = await response.text();
 		} else try {
@@ -14589,7 +14589,7 @@ async function fetchDataset(limit) {
 		logger.debug(`[XSTest] Fetching dataset from ${DATASET_URL}`);
 		let csvData;
 		if (DATASET_URL.startsWith("http")) {
-			const response = await fetchWithTimeout(DATASET_URL, {}, REQUEST_TIMEOUT_MS);
+			const response = await fetchWithTimeout(DATASET_URL, {}, getRequestTimeoutMs());
 			if (!response.ok) throw new Error(`[XSTest] HTTP status: ${response.status} ${response.statusText}`);
 			csvData = await response.text();
 		} else try {
@@ -14867,11 +14867,12 @@ async function fetchRemoteTestCases(key, purpose, injectVar, n, config) {
 			method: "POST",
 			headers: { "Content-Type": "application/json" },
 			body
-		}, REQUEST_TIMEOUT_MS);
+		}, getRequestTimeoutMs());
 		if (status !== 200 || !data || !data.result || !Array.isArray(data.result)) {
 			logger.error(`Error generating test cases for ${key}: ${statusText} ${JSON.stringify(data)}`);
 			return [];
 		}
+		if (requiresRemoteMaterialization(config?.inputs)) assertRemoteMaterializationHandled(data, `Remote plugin generation for ${key}`);
 		const ret = data.result;
 		logger.debug(`Received remote generation for ${key}:\n${JSON.stringify(ret)}`);
 		return ret;
@@ -14880,31 +14881,6 @@ async function fetchRemoteTestCases(key, purpose, injectVar, n, config) {
 		return [];
 	}
 }
-async function materializeRemoteTestCaseInputs({ config, injectVar, pluginId, provider, purpose, testCases }) {
-	const inputs = config.inputs;
-	if (!inputs || Object.keys(inputs).length === 0) return testCases;
-	return Promise.all(testCases.map(async (testCase, materializationIndex) => {
-		const inputVars = extractInputVarsFromPrompt(String(testCase.vars?.[injectVar] ?? ""), inputs);
-		if (!inputVars) return testCase;
-		const materializedVars = await materializeInputVariablesWithMetadata(inputVars, inputs, {
-			materializationIndex,
-			pluginId,
-			provider,
-			purpose
-		});
-		return {
-			...testCase,
-			vars: {
-				...testCase.vars || {},
-				...materializedVars.vars
-			},
-			metadata: {
-				...testCase.metadata || {},
-				...materializedVars.metadata ? { inputMaterialization: materializedVars.metadata } : {}
-			}
-		};
-	}));
-}
 function createPluginFactory(PluginClass, key, validate) {
 	return {
 		key,
@@ -14916,14 +14892,7 @@ function createPluginFactory(PluginClass, key, validate) {
 				return new PluginClass(provider, purpose, injectVar, configWithDefaults).generateTests(n, delayMs);
 			}
 			const pluginId = getShortPluginId(key);
-			const testCases = await materializeRemoteTestCaseInputs({
-				config: configWithDefaults ?? {},
-				injectVar,
-				pluginId,
-				provider,
-				purpose,
-				testCases: await fetchRemoteTestCases(key, purpose, injectVar, n, configWithDefaults ?? {})
-			});
+			const testCases = await fetchRemoteTestCases(key, purpose, injectVar, n, configWithDefaults ?? {});
 			const computedModifiers = computeModifiersFromConfig(configWithDefaults);
 			return testCases.map((testCase) => ({
 				...testCase,
@@ -15010,14 +14979,7 @@ const piiPlugins = PII_PLUGINS.map((category) => ({
 	action: async (params) => {
 		if (shouldGenerateRemote()) {
 			const pluginId = getShortPluginId(category);
-			const testCases = await materializeRemoteTestCaseInputs({
-				config: params.config ?? {},
-				injectVar: params.injectVar,
-				pluginId,
-				provider: params.provider,
-				purpose: params.purpose,
-				testCases: await fetchRemoteTestCases(category, params.purpose, params.injectVar, params.n, params.config ?? {})
-			});
+			const testCases = await fetchRemoteTestCases(category, params.purpose, params.injectVar, params.n, params.config ?? {});
 			const computedModifiers = computeModifiersFromConfig(params.config);
 			return testCases.map((testCase) => ({
 				...testCase,
@@ -15049,14 +15011,7 @@ const biasPlugins = BIAS_PLUGINS.map((category) => ({
 			return [];
 		}
 		const pluginId = getShortPluginId(category);
-		const testCases = await materializeRemoteTestCaseInputs({
-			config: params.config ?? {},
-			injectVar: params.injectVar,
-			pluginId,
-			provider: params.provider,
-			purpose: params.purpose,
-			testCases: await fetchRemoteTestCases(category, params.purpose, params.injectVar, params.n, params.config ?? {})
-		});
+		const testCases = await fetchRemoteTestCases(category, params.purpose, params.injectVar, params.n, params.config ?? {});
 		const computedModifiers = computeModifiersFromConfig(params.config);
 		return testCases.map((testCase) => ({
 			...testCase,
@@ -15075,21 +15030,14 @@ function createRemotePlugin(key, validate) {
 	return {
 		key,
 		validate,
-		action: async ({ provider, purpose, injectVar, n, config }) => {
+		action: async ({ purpose, injectVar, n, config }) => {
 			const configWithDefaults = applyDefaultRemotePluginConfig(key, config);
 			if (neverGenerateRemote()) {
 				logger.error(getRemoteGenerationExplicitlyDisabledError(`${key} plugin`));
 				return [];
 			}
 			const pluginId = getShortPluginId(key);
-			const testCases = await materializeRemoteTestCaseInputs({
-				config: configWithDefaults ?? {},
-				injectVar,
-				pluginId,
-				provider,
-				purpose,
-				testCases: await fetchRemoteTestCases(key, purpose, injectVar, n, configWithDefaults ?? {})
-			});
+			const testCases = await fetchRemoteTestCases(key, purpose, injectVar, n, configWithDefaults ?? {});
 			const computedModifiers = computeModifiersFromConfig(configWithDefaults);
 			const testsWithMetadata = testCases.map((testCase) => ({
 				...testCase,
@@ -15264,7 +15212,7 @@ async function generateCitations(testCases, injectVar, config) {
 				method: "POST",
 				headers: { "Content-Type": "application/json" },
 				body: JSON.stringify(payload)
-			}, REQUEST_TIMEOUT_MS);
+			}, getRequestTimeoutMs());
 			logger.debug(`Got remote citation generation result for case ${Number(index) + 1}: ${JSON.stringify(data)}`);
 			if (data.error) {
 				logger.error(`[Citation] Error in citation generation: ${data.error}`);
@@ -15412,7 +15360,7 @@ async function generateGcgPrompts(testCases, injectVar, config) {
 					"x-promptfoo-silent": "true"
 				},
 				body: JSON.stringify(payload)
-			}, REQUEST_TIMEOUT_MS, "json", true);
+			}, getRequestTimeoutMs(), "json", true);
 			logger.debug("[GCG] Got generation result", {
 				caseNumber,
 				hasError: Boolean(data.error),
@@ -15974,7 +15922,7 @@ async function generateLikertPrompts(testCases, injectVar, config) {
 				method: "POST",
 				headers: { "Content-Type": "application/json" },
 				body: JSON.stringify(payload)
-			}, REQUEST_TIMEOUT_MS);
+			}, getRequestTimeoutMs());
 			logger.debug(`Got Likert jailbreak generation result for case ${Number(index) + 1}: ${JSON.stringify(data)}`);
 			if (data.error || !data.modifiedPrompts) {
 				logger.error(`[jailbreak:likert] Error in Likert generation: ${data.error}}`);
@@ -16059,7 +16007,7 @@ async function generateMathPrompt(testCases, injectVar, config) {
 				method: "POST",
 				headers: { "Content-Type": "application/json" },
 				body: JSON.stringify(payload)
-			}, REQUEST_TIMEOUT_MS);
+			}, getRequestTimeoutMs());
 			logger.debug(`Got remote MathPrompt generation result for batch ${Number(index) + 1}: ${JSON.stringify(data)}`);
 			allResults = allResults.concat(data.result);
 			processedBatches++;
@@ -16643,7 +16591,7 @@ async function textToAudio(text, language = "en", options) {
 			method: "POST",
 			headers: { "Content-Type": "application/json" },
 			body: JSON.stringify(payload)
-		}, REQUEST_TIMEOUT_MS);
+		}, getRequestTimeoutMs());
 		if (data.error || !data.audioBase64) throw new Error(`Error in remote audio generation: ${data.error || "No audio data returned"}`);
 		logger.debug(`Received audio base64 from remote API (${data.audioBase64.length} chars)`);
 		const base64Audio = data.audioBase64;
@@ -17073,7 +17021,7 @@ async function generateCompositePrompts(testCases, injectVar, config) {
 				method: "POST",
 				headers: { "Content-Type": "application/json" },
 				body: JSON.stringify(payload)
-			}, REQUEST_TIMEOUT_MS);
+			}, getRequestTimeoutMs());
 			logger.debug(`Got composite jailbreak generation result for case ${Number(index) + 1}: ${JSON.stringify(data)}`);
 			if (data.error || !data.modifiedPrompts) {
 				logger.error(`[jailbreak:composite] Error in composite generation: ${data.error}}`);
@@ -17462,6 +17410,15 @@ async function loadStrategy(strategyPath) {
 }
 //#endregion
 //#region src/redteam/index.ts
+const MATERIALIZED_MULTI_INPUT_PROMPT_METADATA_KEY = "__promptfooMaterializedMultiInputPrompt";
+function getMaterializedMultiInputPromptSnapshot(metadata) {
+	const snapshot = metadata?.[MATERIALIZED_MULTI_INPUT_PROMPT_METADATA_KEY];
+	return typeof snapshot === "string" ? snapshot : void 0;
+}
+function getMaterializedMultiInputPromptMetadata(vars) {
+	const prompt = vars?.[MULTI_INPUT_VAR];
+	return typeof prompt === "string" ? { [MATERIALIZED_MULTI_INPUT_PROMPT_METADATA_KEY]: prompt } : void 0;
+}
 function getPolicyText(metadata) {
 	if (!metadata || metadata.policy === void 0 || metadata.policy === null) return;
 	const policyValue = metadata.policy;
@@ -17474,12 +17431,19 @@ function getPolicyText(metadata) {
 async function rematerializeStrategyInputVars(testCase, injectVar, provider, purpose, materializationIndex) {
 	const inputs = testCase.metadata?.pluginConfig?.inputs;
 	const inputMaterialization = testCase.metadata?.inputMaterialization;
-	if (!inputs || Object.keys(inputs).length === 0 || !testCase.vars?.[injectVar]) return {
+	const materializedPromptSnapshot = getMaterializedMultiInputPromptSnapshot(testCase.metadata);
+	const currentInjectVar = testCase.vars?.[injectVar];
+	if (!inputs || Object.keys(inputs).length === 0 || !currentInjectVar) return {
+		inputMaterialization,
+		vars: testCase.vars
+	};
+	const promptChangedSinceMaterialization = typeof currentInjectVar === "string" && typeof materializedPromptSnapshot === "string" && currentInjectVar !== materializedPromptSnapshot;
+	if (Boolean(inputMaterialization) && Object.keys(inputs).every((key) => Object.prototype.hasOwnProperty.call(testCase.vars ?? {}, key)) && !promptChangedSinceMaterialization) return {
 		inputMaterialization,
 		vars: testCase.vars
 	};
 	try {
-		const materializedVars = await extractMaterializedVariablesFromJsonWithMetadata(JSON.parse(String(testCase.vars[injectVar])), inputs, {
+		const materializedVars = await extractMaterializedVariablesFromJsonWithMetadata(JSON.parse(String(currentInjectVar)), inputs, {
 			materializationIndex,
 			pluginId: String(testCase.metadata?.pluginId || "unknown-plugin"),
 			provider,
@@ -17711,7 +17675,8 @@ function addLanguageToPluginMetadata(test, lang, plugin, maxCharsPerMessage, tes
 				...test.metadata?.modifiers,
 				...languageToAdd
 			},
-			...languageToAdd
+			...languageToAdd,
+			...getMaterializedMultiInputPromptMetadata(test.vars)
 		}
 	};
 }
@@ -17810,7 +17775,8 @@ async function applyStrategies(testCases, strategies, injectVar, provider, purpo
 					...t?.metadata?.pluginId && { pluginId: t.metadata.pluginId },
 					...t?.metadata?.pluginConfig && { pluginConfig: t.metadata.pluginConfig },
 					...inputMaterialization && { inputMaterialization },
-					...Object.keys(strategyConfig).length > 0 && { strategyConfig }
+					...Object.keys(strategyConfig).length > 0 && { strategyConfig },
+					...getMaterializedMultiInputPromptMetadata(vars)
 				}
 			};
 		})));
@@ -18156,14 +18122,21 @@ async function synthesize({ abortSignal, delay, entities: entitiesOverride, inje
 				}
 			});
 			const languageResults = await Promise.allSettled(languagePromises);
-			for (const result of languageResults) if (result.status === "fulfilled") {
+			for (const [index, result] of languageResults.entries()) if (result.status === "fulfilled") {
 				const { lang, tests, requested, generated } = result.value;
 				allPluginTests.push(...tests);
 				resultsPerLanguage[lang || "default"] = {
 					requested,
 					generated
 				};
-			} else logger.warn(`[Language Processing] Error generating tests for ${plugin.id}: ${result.reason}`);
+			} else {
+				const lang = languages[index];
+				logger.warn(`[Language Processing] Error generating tests for ${plugin.id}: ${result.reason}`);
+				resultsPerLanguage[lang || "default"] = {
+					requested: plugin.numTests,
+					generated: 0
+				};
+			}
 			logger.debug(`[Language Processing] Total tests generated for ${plugin.id}: ${allPluginTests.length} (across ${languages.length} language(s))`);
 			if (!Array.isArray(allPluginTests) || allPluginTests.length === 0) logger.warn(`Failed to generate tests for ${plugin.id}`);
 			else {
@@ -18197,35 +18170,51 @@ async function synthesize({ abortSignal, delay, entities: entitiesOverride, inje
 				generated: allPluginTests.length
 			};
 		} else if (plugin.id.startsWith("file://")) try {
-			const customTests = await new CustomPlugin(redteamProvider, purpose, injectVar, plugin.id, resolvePluginConfigWithMaxChars(plugin.config, maxCharsPerMessage)).generateTests(plugin.numTests, delay);
-			const testCasesWithMetadata = filterOversizedTestCases(customTests.map((t) => {
-				const includePluginConfig = !(t.metadata && Object.hasOwn(t.metadata, "pluginConfig") && t.metadata.pluginConfig === void 0);
-				const pluginConfigWithMaxChars = {
+			const languageConfig = plugin.config?.language ?? language;
+			const languages = Array.isArray(languageConfig) ? languageConfig : languageConfig ? [languageConfig] : [void 0];
+			const allCustomTests = [];
+			const resultsPerLanguage = {};
+			const languagePromises = languages.map(async (lang) => {
+				const resolvedConfig = {
 					...resolvePluginConfigWithMaxChars(plugin.config, maxCharsPerMessage),
-					...t.metadata?.pluginConfig ?? {}
+					...lang ? { language: lang } : {},
+					...hasMultipleInputs ? { inputs } : {}
 				};
-				const modifiers = {
-					...buildRedteamModifiers({
+				const customPluginConfig = {
+					...resolvedConfig,
+					modifiers: buildRedteamModifiers({
 						maxCharsPerMessage,
-						pluginConfig: pluginConfigWithMaxChars,
+						pluginConfig: resolvedConfig,
 						testGenerationInstructions
-					}),
-					...t.metadata?.modifiers
+					})
 				};
+				const testCasesWithMetadata = filterOversizedTestCases((await new CustomPlugin(redteamProvider, purpose, injectVar, plugin.id, customPluginConfig).generateTests(plugin.numTests, delay)).map((t) => addLanguageToPluginMetadata(t, lang, plugin, maxCharsPerMessage, testGenerationInstructions)), injectVar, `Custom plugin ${plugin.id}`, maxCharsPerMessage);
 				return {
-					...t,
-					metadata: {
-						...t.metadata || {},
-						pluginId: plugin.id,
-						...includePluginConfig && { pluginConfig: pluginConfigWithMaxChars },
-						severity: plugin.severity || getPluginSeverity(plugin.id, resolvePluginConfig(plugin.config)),
-						modifiers
-					}
+					lang,
+					tests: testCasesWithMetadata,
+					requested: plugin.numTests,
+					generated: testCasesWithMetadata.length
 				};
-			}), injectVar, `Custom plugin ${plugin.id}`, maxCharsPerMessage);
+			});
+			const languageResults = await Promise.allSettled(languagePromises);
+			for (const [index, result] of languageResults.entries()) if (result.status === "fulfilled") {
+				const { lang, tests, requested, generated } = result.value;
+				allCustomTests.push(...tests);
+				resultsPerLanguage[lang || "default"] = {
+					requested,
+					generated
+				};
+			} else {
+				const lang = languages[index];
+				logger.warn(`[Language Processing] Error generating tests for custom plugin ${plugin.id}: ${result.reason}`);
+				resultsPerLanguage[lang || "default"] = {
+					requested: plugin.numTests,
+					generated: 0
+				};
+			}
 			if (needsGoalExtraction) {
-				logger.debug(`Extracting goal for ${testCasesWithMetadata.length} custom tests from ${plugin.id}...`);
-				for (const testCase of testCasesWithMetadata) {
+				logger.debug(`Extracting goal for ${allCustomTests.length} custom tests from ${plugin.id}...`);
+				for (const testCase of allCustomTests) {
 					const promptVar = testCase.vars?.[injectVar];
 					const prompt = Array.isArray(promptVar) ? promptVar[0] : String(promptVar);
 					const policy = getPolicyText(testCase.metadata);
@@ -18233,13 +18222,21 @@ async function synthesize({ abortSignal, delay, entities: entitiesOverride, inje
 					testCase.metadata.goal = extractedGoal;
 				}
 			}
-			testCases.push(...testCasesWithMetadata);
-			logger.debug(`Added ${customTests.length} custom test cases from ${plugin.id}`);
-			const displayId = getPluginDisplayId(plugin);
-			pluginResults[displayId] = {
-				requested: plugin.numTests,
-				generated: testCasesWithMetadata.length
+			testCases.push(...allCustomTests);
+			logger.debug(`Added ${allCustomTests.length} custom test cases from ${plugin.id}`);
+			const baseDisplayId = getPluginDisplayId(plugin);
+			if (languages.filter((lang) => lang !== void 0).length > 1) for (const [langKey, result] of Object.entries(resultsPerLanguage)) {
+				const displayId = langKey === "en" ? baseDisplayId : `(${langKey}) ${baseDisplayId}`;
+				pluginResults[displayId] = {
+					requested: result.requested,
+					generated: result.generated
+				};
+			}
+			else pluginResults[baseDisplayId] = {
+				requested: plugin.numTests * languages.length,
+				generated: allCustomTests.length
 			};
+			progressBar?.increment(plugin.numTests * languages.length);
 		} catch (e) {
 			logger.error(`Error generating tests for custom plugin ${plugin.id}: ${e}`);
 			const displayId = getPluginDisplayId(plugin);
@@ -19793,7 +19790,7 @@ async function doEval(cmdObj, defaultConfig, defaultConfigPath, evaluateOptions)
 		const passRate = successes / totalTests * 100;
 		if (cmdObj.table && getLogLevel() !== "debug" && totalTests < 500) {
 			const table = await evalRecord.getTable();
-			const outputTable = generateTable(table);
+			const outputTable = generateTable(table, cmdObj.tableCellMaxLength ?? commandLineOptions?.tableCellMaxLength);
 			logger.info("\n" + outputTable.toString());
 			if (table.body.length > 25) {
 				const rowsLeft = table.body.length - 25;
@@ -23267,7 +23264,7 @@ async function handleGoatStrategy(ctx) {
 		method: "POST",
 		headers: { "Content-Type": "application/json" },
 		body: JSON.stringify(goatBody)
-	}, REQUEST_TIMEOUT_MS);
+	}, getRequestTimeoutMs());
 	if (!response.ok) throw new Error(`GOAT task failed with status ${response.status}: ${await response.text()}`);
 	const data = await response.json();
 	const attackerMessage = data?.message;
@@ -23298,7 +23295,7 @@ async function handleMischievousUserStrategy(ctx) {
 		method: "POST",
 		headers: { "Content-Type": "application/json" },
 		body: JSON.stringify(mischievousBody)
-	}, REQUEST_TIMEOUT_MS);
+	}, getRequestTimeoutMs());
 	if (!response.ok) throw new Error(`Mischievous User task failed with status ${response.status}: ${await response.text()}`);
 	const data = await response.json();
 	const result = data?.result;
@@ -23355,7 +23352,7 @@ async function handleHydraStrategy(ctx) {
 		method: "POST",
 		headers: { "Content-Type": "application/json" },
 		body: JSON.stringify(hydraBody)
-	}, REQUEST_TIMEOUT_MS);
+	}, getRequestTimeoutMs());
 	if (!response.ok) throw new Error(`Hydra task failed with status ${response.status}: ${await response.text()}`);
 	const data = await response.json();
 	const rawResult = data?.result;
@@ -23406,7 +23403,7 @@ async function handleCrescendoLikeStrategy(ctx) {
 		method: "POST",
 		headers: { "Content-Type": "application/json" },
 		body: JSON.stringify(providerRequest)
-	}, REQUEST_TIMEOUT_MS);
+	}, getRequestTimeoutMs());
 	if (!response.ok) throw new Error(`Crescendo task failed with status ${response.status}: ${await response.text()}`);
 	const data = await response.json();
 	const rawResult = data?.result;
@@ -24054,7 +24051,7 @@ router.get("/", async (_req, res) => {
 				};
 			} catch (error) {
 				logger.debug(`Failed to fetch latest version: ${error}`);
-				latestVersion = versionCache.latestVersion ?? "0.121.7";
+				latestVersion = versionCache.latestVersion ?? "0.121.9";
 			}
 		}
 		const selfHosted = getEnvBool("PROMPTFOO_SELF_HOSTED");
@@ -24063,7 +24060,7 @@ router.get("/", async (_req, res) => {
 			selfHosted,
 			isNpx
 		});
-		const resolvedLatestVersion = latestVersion ?? "0.121.7";
+		const resolvedLatestVersion = latestVersion ?? "0.121.9";
 		const response = {
 			currentVersion: VERSION,
 			latestVersion: resolvedLatestVersion,

package/dist/src/{server-DhMHosWj.js → server-4wx571OM.js}

@@ -1,8 +1,8 @@
 #!/usr/bin/env node
-import { s as logger } from "./logger-BbY6ypFL.js";
-import { N as getDefaultPort, n as fetchWithProxy } from "./fetch-DXUnXkVU.js";
-import "./version-eRkNuGv8.js";
-import { o as getRemoteVersionUrl } from "./remoteGeneration-B1_XsKXU.js";
+import { s as logger } from "./logger-h-eiS4iv.js";
+import { N as getDefaultPort, n as fetchWithProxy } from "./fetch-C59QM4cf.js";
+import "./version-CnYqmWbz.js";
+import { o as getRemoteVersionUrl } from "./remoteGeneration-DW6mPZnO.js";
 import chalk from "chalk";
 import opener from "opener";
 import readline from "readline";
@@ -109,7 +109,7 @@ async function checkServerRunning(port = getDefaultPort()) {
 	logger.debug(`Checking for existing server on port ${port}...`);
 	try {
 		const data = await (await fetchWithProxy(`http://localhost:${port}/health`, { headers: { "x-promptfoo-silent": "true" } })).json();
-		return data.status === "OK" && data.version === "0.121.7";
+		return data.status === "OK" && data.version === "0.121.9";
 	} catch (err) {
 		logger.debug(`No existing server found - this is expected on first startup. ${String(err)}`);
 		return false;
@@ -179,4 +179,4 @@ async function openAuthBrowser(authUrl, welcomeUrl, browserBehavior) {
 //#endregion
 export { openAuthBrowser as a, promptYesNo as c, checkServerRunning as i, BrowserBehaviorNames as n, openBrowser as o, checkServerFeatureSupport as r, promptUser as s, BrowserBehavior as t };
 
-//# sourceMappingURL=server-DhMHosWj.js.map
+//# sourceMappingURL=server-4wx571OM.js.map