openfox 2.0.117 → 2.0.119

package/dist/{chunk-GZTMCTUS.js → chunk-U6JMRMBW.js}

@@ -1,6 +1,9 @@
+import {
+  parseLmStudioModels
+} from "./chunk-K2QKX4ZJ.js";
 import {
   createLLMClient
-} from "./chunk-MKWS7TQH.js";
+} from "./chunk-J6KFWVZ3.js";
 import {
   getModelProfile
 } from "./chunk-W7VIYHPQ.js";
@@ -14,7 +17,7 @@ import {
 } from "./chunk-HNCM3D7Y.js";
 import {
   resolveAttachmentsInMessages
-} from "./chunk-TOKQ2UVG.js";
+} from "./chunk-MIPQOHWS.js";
 import {
   logger
 } from "./chunk-K44MW7JJ.js";
@@ -328,29 +331,18 @@ async function fetchLmStudioModelsWithContext(baseUrl, _apiKey) {
       logger.info("LM Studio native endpoint returned error", { url: nativeUrl, status: response.status });
       return [];
     }
-    const raw = await response.json();
-    const rawArray = Array.isArray(raw) ? raw : Array.isArray(raw.models) ? raw.models : Array.isArray(raw.data) ? raw.data : [];
-    const modelList = rawArray;
+    const modelList = parseLmStudioModels(await response.json());
     if (modelList.length === 0) {
       logger.info("LM Studio native endpoint returned no models", { url: nativeUrl });
       return [];
     }
     logger.info("LM Studio native models found", { count: modelList.length });
     return modelList.map((model) => {
-      const modelId = model.key ?? model.id ?? "";
-      const loadedContext = model.loaded_instances?.[0]?.config?.context_length;
-      const maxContext = model.max_context_length;
-      const contextWindow = loadedContext ?? maxContext ?? 2e5;
-      logger.info("LM Studio model detected", {
-        modelId,
-        loadedContext,
-        maxContext,
-        contextWindow
-      });
+      logger.info("LM Studio model detected", { modelId: model.id, contextWindow: model.contextWindow });
       return {
-        id: modelId,
-        contextWindow,
-        source: loadedContext || maxContext ? "backend" : "default"
+        id: model.id,
+        contextWindow: model.contextWindow ?? 2e5,
+        source: model.contextWindow ? "backend" : "default"
       };
     });
   } catch (error) {
@@ -703,7 +695,8 @@ function createProviderManager(config, options = {}) {
         ...settings.thinkingExtraKwargs !== void 0 ? { thinkingExtraKwargs: settings.thinkingExtraKwargs } : existingModel?.thinkingExtraKwargs !== void 0 ? { thinkingExtraKwargs: existingModel.thinkingExtraKwargs } : {},
         ...settings.nonThinkingExtraKwargs !== void 0 ? { nonThinkingExtraKwargs: settings.nonThinkingExtraKwargs } : existingModel?.nonThinkingExtraKwargs !== void 0 ? { nonThinkingExtraKwargs: existingModel.nonThinkingExtraKwargs } : {},
         ...settings.thinkingQueryParams !== void 0 ? { thinkingQueryParams: settings.thinkingQueryParams } : existingModel?.thinkingQueryParams !== void 0 ? { thinkingQueryParams: existingModel.thinkingQueryParams } : {},
-        ...settings.nonThinkingQueryParams !== void 0 ? { nonThinkingQueryParams: settings.nonThinkingQueryParams } : existingModel?.nonThinkingQueryParams !== void 0 ? { nonThinkingQueryParams: existingModel.nonThinkingQueryParams } : {}
+        ...settings.nonThinkingQueryParams !== void 0 ? { nonThinkingQueryParams: settings.nonThinkingQueryParams } : existingModel?.nonThinkingQueryParams !== void 0 ? { nonThinkingQueryParams: existingModel.nonThinkingQueryParams } : {},
+        ...settings.omitParams !== void 0 ? { omitParams: settings.omitParams } : existingModel?.omitParams !== void 0 ? { omitParams: existingModel.omitParams } : {}
       });
       if (existingModel) {
         providers = providers.map(
@@ -725,6 +718,7 @@ function createProviderManager(config, options = {}) {
       if (model["topK"] !== void 0) baseSettings["topK"] = model["topK"];
       if (model["maxTokens"] !== void 0) baseSettings["maxTokens"] = model["maxTokens"];
       if (model["supportsVision"] !== void 0) baseSettings["supportsVision"] = model["supportsVision"];
+      if (model["omitParams"] !== void 0) baseSettings["omitParams"] = model["omitParams"];
       const rawQueryParams = mode === "thinking" ? model.thinkingQueryParams : model.nonThinkingQueryParams;
       if (rawQueryParams) {
         return { ...baseSettings, queryParams: JSON.parse(rawQueryParams) };
@@ -740,6 +734,7 @@ function createProviderManager(config, options = {}) {
       if (fallbackRawQP) {
         return { ...baseSettings, queryParams: JSON.parse(fallbackRawQP) };
       }
+      if (model["omitParams"] !== void 0) return baseSettings;
       return void 0;
     },
     async refreshProviderModels(providerId) {
@@ -762,8 +757,8 @@ function createProviderManager(config, options = {}) {
         backendModelsCount: modelsWithContext.length
       });
       if (modelsWithContext.length === 0) {
-        providerStatus.set(providerId, "disconnected");
         if (userModels.length > 0) {
+          providerStatus.set(providerId, "unknown");
           logger.debug("Backend unavailable, preserving user models", {
             providerId,
             userModels: userModels.map((m) => ({ id: m.id, contextWindow: m.contextWindow }))
@@ -771,6 +766,7 @@ function createProviderManager(config, options = {}) {
           providers = providers.map((p) => p.id === providerId ? { ...p, models: userModels } : p);
           return { success: true };
         }
+        providerStatus.set(providerId, "disconnected");
         return { success: false, error: "No models returned from backend" };
       }
       providerStatus.set(providerId, "connected");
@@ -803,4 +799,4 @@ export {
   parseDefaultModelSelection,
   createProviderManager
 };
-//# sourceMappingURL=chunk-GZTMCTUS.js.map
+//# sourceMappingURL=chunk-U6JMRMBW.js.map

package/dist/{chunk-3PEYTWCS.js → chunk-UNNCD7A3.js}

@@ -105,7 +105,7 @@ function resolveAgentDef(sessionManager, sessionId) {
 }
 async function buildCachedPrompt(sessionManager, sessionId, agentDef, modelName) {
   const { instructionContent, skills } = await loadSessionContext(sessionManager, sessionId);
-  const { getToolRegistryForAgent } = await import("./tools-2Q3GRECI.js");
+  const { getToolRegistryForAgent } = await import("./tools-ZJXDZLRH.js");
   const tools = getToolRegistryForAgent(agentDef, sessionId).definitions;
   const toolFingerprint = getToolFingerprint(tools);
   const allAgents = await loadAllAgentsDefault(sessionManager.getProjectWorkdir(sessionId));
@@ -124,7 +124,7 @@ async function buildCachedPrompt(sessionManager, sessionId, agentDef, modelName)
 async function computeSessionHash(sessionManager, sessionId, modelName) {
   const { instructionContent, skills } = await loadSessionContext(sessionManager, sessionId);
   const agentDef = await resolveAgentDef(sessionManager, sessionId);
-  const { getToolRegistryForAgent } = await import("./tools-2Q3GRECI.js");
+  const { getToolRegistryForAgent } = await import("./tools-ZJXDZLRH.js");
   const tools = getToolRegistryForAgent(agentDef, sessionId).definitions;
   const toolFingerprint = getToolFingerprint(tools);
   return computeDynamicContextHash(instructionContent, skills, toolFingerprint, modelName);
@@ -148,4 +148,4 @@ export {
   computeSessionHash,
   applyDynamicContext
 };
-//# sourceMappingURL=chunk-3PEYTWCS.js.map
+//# sourceMappingURL=chunk-UNNCD7A3.js.map

package/dist/{chunk-2PQA3DHY.js → chunk-V5XWUVF5.js}

@@ -7,14 +7,18 @@ import {
 } from "./chunk-2FCLCUVR.js";
 import {
   TurnMetrics,
+  clearLLMFailure,
   consumeStreamGenerator,
   createChatDoneEvent,
   createMessageDoneEvent,
   createMessageStartEvent,
   createToolCallEvent,
   createToolResultEvent,
+  evaluateLLMRetry,
+  recordLLMFailure,
+  sleepThroughRetryBackoff,
   streamLLMPure
-} from "./chunk-EQ7DK2TG.js";
+} from "./chunk-SE62ZBNX.js";
 import {
   COMPACTION_PROMPT,
   buildBasePrompt,
@@ -57,7 +61,7 @@ import {
   getGateConfig,
   isTaskConflictError,
   isTaskGateError
-} from "./chunk-HZSKKJHQ.js";
+} from "./chunk-5FAPM2GM.js";
 import {
   getSessionDisabledServers
 } from "./chunk-EQ5CGNOT.js";
@@ -67,19 +71,21 @@ import {
   extractSensitivePathsFromCommand,
   registerPathConfirmation,
   requestPathAccess
-} from "./chunk-QQVPUZQB.js";
+} from "./chunk-XZGQZ42D.js";
 import {
   getCurrentContextWindowId,
   getCurrentWindowMessageOptions,
   getEventStore
-} from "./chunk-TGVDUK4N.js";
+} from "./chunk-JOCR4CF4.js";
 import {
   createChatDoneMessage,
+  createChatLLMRetryFailedMessage,
+  createChatLLMRetryMessage,
   createChatMessageMessage,
   createChatMessageUpdatedMessage,
   createChatPathConfirmationMessage,
   createQueueStateMessage
-} from "./chunk-4GHLULHB.js";
+} from "./chunk-XAAEP5XM.js";
 import {
   AskUserInterrupt,
   askUserTool
@@ -116,7 +122,7 @@ import {
   formatPdfErrorMessage,
   isPdfBuffer,
   processPdfContent
-} from "./chunk-TOKQ2UVG.js";
+} from "./chunk-MIPQOHWS.js";
 import {
   SETTINGS_KEYS,
   getSetting
@@ -1760,6 +1766,14 @@ function drainQueue(sessionManager, sessionId, append, onMessage) {
   return { messages: asapMessages, hasMessages: asapMessages.length > 0 };
 }
 
+// src/server/runner/types.ts
+var DEFAULT_LLM_RETRY_POLICY = {
+  backoffMs: [1e3, 5e3, 3e4],
+  minIntervalMs: 6e4,
+  maxDurationMs: 30 * 6e4,
+  maxAttempts: 40
+};
+
 // src/server/chat/agent-loop.ts
 function emitPartialDoneEvents(_sessionId, assistantMsgId, statsIdentity, mode, turnMetrics, append, agentType) {
   const stats = turnMetrics.buildStats(statsIdentity, mode);
@@ -1791,11 +1805,14 @@ function emitDoneAndBreak(assistantMsgId, segments, statsIdentity, mode, turnMet
   }
 }
 var MAX_TRUNCATION_RETRIES = 3;
+var OUTPUT_RESERVE_TOKENS = 2048;
 var CONTINUE_PROMPT = "Continue your previous response. Do NOT repeat what you already wrote.";
+var CONTINUE_AFTER_STREAM_ERROR_PROMPT = "The LLM stream was interrupted mid-response. Continue exactly where you left off \u2014 do not repeat what was already written.";
 async function runTopLevelAgentLoop(config, turnMetrics) {
   const { mode, sessionManager, sessionId, llmClient, signal, onMessage, statsIdentity } = config;
   const append = config.append;
   const agentType = config.subAgentMetadata ? "sub-agent" : void 0;
+  const resolveClient = () => config.getLLMClient?.() ?? llmClient;
   const retryLimiter = createRetryLimiter(config.maxRetriesPerTurn ?? 10);
   let truncationRetryCount = 0;
   let returnValueContent;
@@ -1809,26 +1826,26 @@ async function runTopLevelAgentLoop(config, turnMetrics) {
     if (config.warmup) {
       const session2 = sessionManager.requireSession(sessionId);
       const runtimeConfig2 = getRuntimeConfig();
-      const configDir2 = getGlobalConfigDir(runtimeConfig2.mode ?? "production");
-      const skills2 = await getEnabledSkillMetadata(configDir2, sessionManager.getProjectWorkdir(sessionId));
+      const configDir = getGlobalConfigDir(runtimeConfig2.mode ?? "production");
+      const skills = await getEnabledSkillMetadata(configDir, sessionManager.getProjectWorkdir(sessionId));
       const { content: instructionContent2 } = await getAllInstructions(session2.workdir, session2.projectId);
       const toolRegistry2 = config.getToolRegistry();
-      const assembledRequest2 = await config.assembleRequest({
+      const assembledRequest = await config.assembleRequest({
         workdir: session2.workdir,
         messages: [],
         injectedFiles: [],
         promptTools: toolRegistry2.definitions,
         toolChoice: "none",
         ...instructionContent2 ? { customInstructions: instructionContent2 } : {},
-        ...skills2.length > 0 ? { skills: skills2 } : {}
+        ...skills.length > 0 ? { skills } : {}
       });
-      const modelSettings2 = sessionManager.getCurrentModelSettings(sessionId);
-      await llmClient.complete({
-        messages: [{ role: "system", content: assembledRequest2.systemPrompt }],
-        tools: assembledRequest2.tools,
+      const modelSettings = sessionManager.getCurrentModelSettings(sessionId);
+      await resolveClient().complete({
+        messages: [{ role: "system", content: assembledRequest.systemPrompt }],
+        tools: assembledRequest.tools,
         maxTokens: 1,
         temperature: 0,
-        ...modelSettings2 ? { modelSettings: modelSettings2 } : {}
+        ...modelSettings ? { modelSettings } : {}
       });
       return {};
     }
@@ -1845,71 +1862,125 @@ async function runTopLevelAgentLoop(config, turnMetrics) {
     }));
     const toolRegistry = config.getToolRegistry();
     const currentWindowMessageOptions = getCurrentWindowMessageOptions(sessionId);
-    const requestMessages = await config.getConversationMessages();
-    if (retryLimiter.count() > 0) {
-      const continueMsgId = crypto.randomUUID();
-      const continueContent = lastPatternMatch ? `Your previous response was interrupted because it matched pattern "${lastPatternMatch.pattern}" in ${lastPatternMatch.field}.
+    const retryPolicy = { ...DEFAULT_LLM_RETRY_POLICY, ...config.llmRetryPolicy };
+    const runtimeConfig = getRuntimeConfig();
+    let requestFailures = 0;
+    let requestFirstFailureAt = 0;
+    let continuationAppended = false;
+    let previousContextTokens;
+    let result;
+    let assistantMsgId;
+    let assistantMessageStarted = false;
+    for (; ; ) {
+      const requestMessages = await config.getConversationMessages();
+      if (requestFailures === 0 && retryLimiter.count() > 0) {
+        const continueMsgId = crypto.randomUUID();
+        const continueContent = lastPatternMatch ? `Your previous response was interrupted because it matched pattern "${lastPatternMatch.pattern}" in ${lastPatternMatch.field}.
 Matched content:
 ${lastPatternMatch.matchedContent}
 
 ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
-      append(
-        createMessageStartEvent(continueMsgId, "user", continueContent, {
-          ...currentWindowMessageOptions ?? {},
-          isSystemGenerated: true,
-          messageKind: "correction"
-        })
-      );
-      append({ type: "message.done", data: { messageId: continueMsgId } });
-      requestMessages.push({ role: "user", content: continueContent, source: "history" });
+        append(
+          createMessageStartEvent(continueMsgId, "user", continueContent, {
+            ...currentWindowMessageOptions ?? {},
+            isSystemGenerated: true,
+            messageKind: "correction"
+          })
+        );
+        append({ type: "message.done", data: { messageId: continueMsgId } });
+        requestMessages.push({ role: "user", content: continueContent, source: "history" });
+      }
+      const configDir = getGlobalConfigDir(runtimeConfig.mode ?? "production");
+      const skills = await getEnabledSkillMetadata(configDir, sessionManager.getProjectWorkdir(sessionId));
+      if (signal?.aborted) throw new Error("Aborted");
+      const assembledRequest = await config.assembleRequest({
+        workdir: session.workdir,
+        messages: requestMessages,
+        injectedFiles,
+        promptTools: toolRegistry.definitions,
+        toolChoice: "auto",
+        ...instructionContent ? { customInstructions: instructionContent } : {},
+        ...skills.length > 0 ? { skills } : {}
+      });
+      assistantMsgId = crypto.randomUUID();
+      assistantMessageStarted = false;
+      const ensureAssistantMessage = () => {
+        if (assistantMessageStarted) return;
+        assistantMessageStarted = true;
+        append(
+          createMessageStartEvent(assistantMsgId, "assistant", void 0, {
+            ...currentWindowMessageOptions ?? {},
+            ...config.subAgentMetadata ? { subAgentId: config.subAgentMetadata.subAgentId, subAgentType: config.subAgentMetadata.subAgentType } : {}
+          })
+        );
+      };
+      const contextState = sessionManager.getContextState(sessionId);
+      previousContextTokens = contextState.currentTokens;
+      const contextWindow = sessionManager.getCurrentModelContext();
+      const availableForOutput = Math.max(256, contextWindow - contextState.currentTokens - OUTPUT_RESERVE_TOKENS);
+      let modelSettings = config.modelSettings ?? (currentMaxTokensOverride !== void 0 ? { ...sessionManager.getCurrentModelSettings(sessionId), maxTokens: currentMaxTokensOverride } : sessionManager.getCurrentModelSettings(sessionId));
+      if (modelSettings) {
+        const requestedMaxTokens = modelSettings.maxTokens ?? 16384;
+        modelSettings = { ...modelSettings, maxTokens: Math.min(requestedMaxTokens, availableForOutput) };
+      }
+      const allAgents = await loadAllAgentsDefault(sessionManager.getProjectWorkdir(sessionId));
+      const subAgentAliases = new Set(getSubAgents(allAgents).map((a) => a.metadata.id));
+      const streamGen = streamLLMPure({
+        messageId: assistantMsgId,
+        systemPrompt: assembledRequest.systemPrompt,
+        llmClient: resolveClient(),
+        messages: assembledRequest.messages,
+        tools: assembledRequest.tools,
+        toolChoice: "auto",
+        signal,
+        subAgentAliases,
+        ...config.retryPatterns ? { retryPatterns: config.retryPatterns } : {},
+        ...modelSettings && { modelSettings }
+      });
+      const attemptResult = await consumeStreamGenerator(streamGen, (event) => {
+        ensureAssistantMessage();
+        append(event);
+      });
+      if (!attemptResult.error) {
+        result = attemptResult;
+        break;
+      }
+      if (assistantMessageStarted && !continuationAppended) {
+        append(createMessageDoneEvent(assistantMsgId, { partial: true }));
+        onMessage?.(createChatMessageUpdatedMessage(assistantMsgId, { isStreaming: false, partial: true }));
+        const continueMsgId = crypto.randomUUID();
+        append(
+          createMessageStartEvent(continueMsgId, "user", CONTINUE_AFTER_STREAM_ERROR_PROMPT, {
+            ...currentWindowMessageOptions ?? {},
+            isSystemGenerated: true,
+            messageKind: "correction"
+          })
+        );
+        append({ type: "message.done", data: { messageId: continueMsgId } });
+        continuationAppended = true;
+      }
+      if (signal?.aborted) throw new Error("Aborted");
+      requestFailures += 1;
+      if (requestFirstFailureAt === 0) {
+        requestFirstFailureAt = Date.now();
+      }
+      const decision = evaluateLLMRetry(requestFailures, requestFirstFailureAt, Date.now(), retryPolicy);
+      if (!decision.retry) {
+        if (!config.subAgentMetadata) {
+          recordLLMFailure(sessionId);
+          config.onMessage?.(createChatLLMRetryFailedMessage(attemptResult.error, requestFailures));
+        }
+        return { failed: { error: attemptResult.error } };
+      }
+      if (!config.subAgentMetadata) {
+        config.onMessage?.(createChatLLMRetryMessage(decision.attempt, decision.delayMs));
+      }
+      const waitResult = await sleepThroughRetryBackoff(decision.delayMs, sessionId, signal);
+      if (waitResult === "aborted") throw new Error("Aborted");
+    }
+    if (!config.subAgentMetadata) {
+      clearLLMFailure(sessionId);
     }
-    const runtimeConfig = getRuntimeConfig();
-    const configDir = getGlobalConfigDir(runtimeConfig.mode ?? "production");
-    const skills = await getEnabledSkillMetadata(configDir, sessionManager.getProjectWorkdir(sessionId));
-    if (signal?.aborted) throw new Error("Aborted");
-    const assembledRequest = await config.assembleRequest({
-      workdir: session.workdir,
-      messages: requestMessages,
-      injectedFiles,
-      promptTools: toolRegistry.definitions,
-      toolChoice: "auto",
-      ...instructionContent ? { customInstructions: instructionContent } : {},
-      ...skills.length > 0 ? { skills } : {}
-    });
-    const assistantMsgId = crypto.randomUUID();
-    append(
-      createMessageStartEvent(assistantMsgId, "assistant", void 0, {
-        ...currentWindowMessageOptions ?? {},
-        ...config.subAgentMetadata ? { subAgentId: config.subAgentMetadata.subAgentId, subAgentType: config.subAgentMetadata.subAgentType } : {}
-      })
-    );
-    const contextState = sessionManager.getContextState(sessionId);
-    const previousContextTokens = contextState.currentTokens;
-    const contextWindow = sessionManager.getCurrentModelContext();
-    const availableForOutput = Math.max(256, contextWindow - contextState.currentTokens);
-    let modelSettings = config.modelSettings ?? (currentMaxTokensOverride !== void 0 ? { ...sessionManager.getCurrentModelSettings(sessionId), maxTokens: currentMaxTokensOverride } : sessionManager.getCurrentModelSettings(sessionId));
-    if (modelSettings) {
-      const requestedMaxTokens = modelSettings.maxTokens ?? 16384;
-      modelSettings = { ...modelSettings, maxTokens: Math.min(requestedMaxTokens, availableForOutput) };
-    }
-    const allAgents = await loadAllAgentsDefault(sessionManager.getProjectWorkdir(sessionId));
-    const subAgentAliases = new Set(getSubAgents(allAgents).map((a) => a.metadata.id));
-    const streamGen = streamLLMPure({
-      messageId: assistantMsgId,
-      systemPrompt: assembledRequest.systemPrompt,
-      llmClient,
-      messages: assembledRequest.messages,
-      tools: assembledRequest.tools,
-      toolChoice: "auto",
-      signal,
-      subAgentAliases,
-      ...config.retryPatterns ? { retryPatterns: config.retryPatterns } : {},
-      ...config.suppressChatError ? { suppressChatError: true } : {},
-      ...modelSettings && { modelSettings }
-    });
-    const result = await consumeStreamGenerator(streamGen, (event) => {
-      append(event);
-    });
     if (result.patternMatch) {
       if (!retryLimiter.canRetry()) {
         append({
@@ -1949,41 +2020,30 @@ ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
       continue;
     }
     if (result.aborted) {
-      emitPartialDoneEvents(sessionId, assistantMsgId, statsIdentity, mode, turnMetrics, append, agentType);
+      if (assistantMessageStarted) {
+        emitPartialDoneEvents(sessionId, assistantMsgId, statsIdentity, mode, turnMetrics, append, agentType);
+      }
       throw new Error("Aborted");
     }
-    if (result.error) {
-      const stats2 = turnMetrics.buildStats(statsIdentity, mode);
-      append(
-        createMessageDoneEvent(assistantMsgId, {
-          segments: result.segments,
-          stats: stats2
-        })
-      );
-      append(createChatDoneEvent(assistantMsgId, "error", stats2, agentType));
-      return { failed: { error: result.error } };
-    }
-    if (!result.error) {
-      turnMetrics.addLLMCall(
-        result.timing,
-        result.usage.promptTokens,
-        result.usage.completionTokens,
-        previousContextTokens,
-        result.modelParams
-      );
-      sessionManager.setCurrentContextSize(
-        sessionId,
-        result.usage.promptTokens,
-        result.usage.completionTokens,
-        config.subAgentMetadata?.subAgentId
-      );
-    }
+    turnMetrics.addLLMCall(
+      result.timing,
+      result.usage.promptTokens,
+      result.usage.completionTokens,
+      previousContextTokens,
+      result.modelParams
+    );
+    sessionManager.setCurrentContextSize(
+      sessionId,
+      result.usage.promptTokens,
+      result.usage.completionTokens,
+      config.subAgentMetadata?.subAgentId
+    );
     if (!compacting) {
-      const contextState2 = sessionManager.getContextState(sessionId);
-      const { shouldCompact, appendCompactionPrompt } = await import("./compactor-LAXZLPPA.js");
+      const contextState = sessionManager.getContextState(sessionId);
+      const { shouldCompact, appendCompactionPrompt } = await import("./compactor-F6XMYVHL.js");
       if (shouldCompact(
-        contextState2.currentTokens,
-        contextState2.maxTokens,
+        contextState.currentTokens,
+        contextState.maxTokens,
         sessionManager.getModelCompactionThreshold(sessionId) ?? runtimeConfig.context.compactionThreshold
       )) {
         appendCompactionPrompt(sessionId, append);
@@ -1996,8 +2056,11 @@ ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
         truncationRetryCount += 1;
         const currentMaxTokens = result.modelParams?.maxTokens ?? 16384;
         const promptTokens = result.usage.promptTokens;
-        const contextWindow2 = sessionManager.getCurrentModelContext();
-        const newMaxTokens = Math.min(Math.floor(currentMaxTokens * 1.5), contextWindow2 - promptTokens - 2048);
+        const contextWindow = sessionManager.getCurrentModelContext();
+        const newMaxTokens = Math.min(
+          Math.floor(currentMaxTokens * 1.5),
+          contextWindow - promptTokens - OUTPUT_RESERVE_TOKENS
+        );
         currentMaxTokensOverride = newMaxTokens;
         const interimStats = turnMetrics.buildStats(statsIdentity, mode);
         append(
@@ -2069,7 +2132,7 @@ ${COMPACTION_PROMPT}`,
           turnMetrics,
           signal,
           onMessage,
-          llmClient,
+          llmClient: resolveClient(),
           statsIdentity,
           onToolExecuted: config.onToolExecuted
         };
@@ -2970,7 +3033,7 @@ var callSubAgentTool = {
       };
     }
     try {
-      const { getToolRegistryForAgent: getToolRegistryForAgent2 } = await import("./tools-2Q3GRECI.js");
+      const { getToolRegistryForAgent: getToolRegistryForAgent2 } = await import("./tools-ZJXDZLRH.js");
       const toolRegistry = getToolRegistryForAgent2(agentDef);
       const turnMetrics = new TurnMetrics();
       const result = await executeSubAgent({
@@ -3992,7 +4055,7 @@ var mcpConfigTool = createTool(
       mcpNotifyChanged?.(sessionId);
     }
     async function rebuildTools() {
-      const { setMcpTools: setMcpTools2 } = await import("./tools-2Q3GRECI.js");
+      const { setMcpTools: setMcpTools2 } = await import("./tools-ZJXDZLRH.js");
       const mcpTools = createMcpTools(mcpManagerForTools);
       setMcpTools2(mcpTools);
     }
@@ -4876,4 +4939,4 @@ export {
   getToolRegistryForAgent,
   createToolRegistry
 };
-//# sourceMappingURL=chunk-2PQA3DHY.js.map
+//# sourceMappingURL=chunk-V5XWUVF5.js.map

package/dist/{chunk-6XHDNSA3.js → chunk-VFS7TFA7.js}

@@ -1,14 +1,14 @@
 import {
   combineEventsWithSnapshot,
   getEventStore
-} from "./chunk-TGVDUK4N.js";
+} from "./chunk-JOCR4CF4.js";
 import {
   updateSessionMetadata
 } from "./chunk-KLQEL2DZ.js";
 import {
   createContextStateMessage,
   createSessionStateMessage
-} from "./chunk-4GHLULHB.js";
+} from "./chunk-XAAEP5XM.js";
 import {
   getPendingQuestionsForSession
 } from "./chunk-JURM3RPZ.js";
@@ -51,6 +51,7 @@ function buildRunChatTurnParams(params) {
     sessionManager: params.sessionManager,
     sessionId: params.sessionId,
     llmClient: params.llmClient,
+    ...params.getSessionLLMClient ? { getSessionLLMClient: params.getSessionLLMClient } : {},
     signal: params.signal,
     onMessage: params.onMessage,
     ...params.statsIdentity ? { statsIdentity: params.statsIdentity } : {}
@@ -103,6 +104,7 @@ function applyGeneratedSessionName(sessionId, name, deps) {
     const { messages, hiddenCount } = buildMessagesFromStoredEvents(events, maxVisibleItems || void 0);
     const pendingConfirmations = foldPendingConfirmations(events);
     const pendingQuestions = getPendingQuestionsForSession(sessionId);
+    const activeWorkflowExecution = deps.sessionManager.getDisplayWorkflowExecution?.(sessionId) ?? void 0;
     deps.broadcastForSession(
       sessionId,
       createSessionStateMessage(
@@ -112,7 +114,8 @@ function applyGeneratedSessionName(sessionId, name, deps) {
         pendingQuestions,
         void 0,
         void 0,
-        hiddenCount
+        hiddenCount,
+        activeWorkflowExecution
       )
     );
   }
@@ -153,7 +156,7 @@ async function generateSessionNameForSession(sessionId, userMessage, deps, signa
       client = deps.getLLMClient();
       client.setModel(providerConfig.model);
     } else if (providerConfig) {
-      const { createLLMClient } = await import("./client-KCNTDTIR.js");
+      const { createLLMClient } = await import("./client-R4PI64IL.js");
       client = createLLMClient({
         llm: {
           baseUrl: providerConfig.baseUrl,
@@ -204,4 +207,4 @@ export {
   needsNameGeneration,
   generateSessionNameForSession
 };
-//# sourceMappingURL=chunk-6XHDNSA3.js.map
+//# sourceMappingURL=chunk-VFS7TFA7.js.map

package/dist/{chunk-4GHLULHB.js → chunk-XAAEP5XM.js}

@@ -4,7 +4,7 @@ import {
 import {
   createServerMessage,
   isClientMessage
-} from "./chunk-ZKLE2Q3I.js";
+} from "./chunk-KSDWQ6IZ.js";
 
 // src/server/session/client-session.ts
 function toClientSession(session) {
@@ -140,11 +140,11 @@ function createChatDoneMessage(messageId, reason, stats, agentType) {
 function createChatErrorMessage(error, recoverable) {
   return createServerMessage("chat.error", { error, recoverable });
 }
-function createChatMessageRemovedMessage(messageIds) {
-  return createServerMessage("chat.message_removed", { messageIds });
+function createChatLLMRetryMessage(attempt, retryInMs) {
+  return createServerMessage("chat.llm_retry", { attempt, retryInMs });
 }
-function createChatStepRetryMessage(stepName, attempt, retryInMs) {
-  return createServerMessage("chat.step_retry", { stepName, attempt, retryInMs });
+function createChatLLMRetryFailedMessage(error, attempts) {
+  return createServerMessage("chat.llm_retry_failed", { error, attempts });
 }
 function createChatPathConfirmationMessage(callId, tool, paths, workdir, reason) {
   return createServerMessage("chat.path_confirmation", { callId, tool, paths, workdir, reason });
@@ -322,14 +322,6 @@ function storedEventToServerMessage(event) {
       const data = event.data;
       return createChatFormatRetryMessage(data.attempt, data.maxAttempts, data.pattern, data.field, data.matchedContent);
     }
-    case "message.removed": {
-      const data = event.data;
-      return createChatMessageRemovedMessage(data.messageIds);
-    }
-    case "workflow.step_retry": {
-      const data = event.data;
-      return createChatStepRetryMessage(data.stepName, data.attempt, data.retryInMs);
-    }
     case "path.confirmation_pending":
     case "path.confirmation_responded":
       return null;
@@ -365,8 +357,8 @@ export {
   createChatMessageUpdatedMessage,
   createChatDoneMessage,
   createChatErrorMessage,
-  createChatMessageRemovedMessage,
-  createChatStepRetryMessage,
+  createChatLLMRetryMessage,
+  createChatLLMRetryFailedMessage,
   createChatPathConfirmationMessage,
   createChatAskUserMessage,
   createModeChangedMessage,
@@ -382,4 +374,4 @@ export {
   createQueueStateMessage,
   storedEventToServerMessage
 };
-//# sourceMappingURL=chunk-4GHLULHB.js.map
+//# sourceMappingURL=chunk-XAAEP5XM.js.map