openfox 2.0.24 → 2.0.26

package/dist/chunk-J2GP3J3X.js

@@ -0,0 +1,97 @@
+// src/server/llm/streaming.ts
+async function* streamWithSegments(client, request) {
+  let content = "";
+  let thinkingContent = "";
+  let response = null;
+  const segments = [];
+  let currentTextSegment = "";
+  let currentThinkingSegment = "";
+  const startTime = performance.now();
+  let firstTokenTime = null;
+  const flushText = () => {
+    if (currentTextSegment.trim()) {
+      segments.push({ type: "text", content: currentTextSegment });
+    }
+    currentTextSegment = "";
+  };
+  const flushThinking = () => {
+    if (currentThinkingSegment.trim()) {
+      segments.push({ type: "thinking", content: currentThinkingSegment });
+    }
+    currentThinkingSegment = "";
+  };
+  try {
+    for await (const event of client.stream(request)) {
+      switch (event.type) {
+        case "text_delta":
+          if (firstTokenTime === null) {
+            firstTokenTime = performance.now();
+          }
+          flushThinking();
+          content += event.content;
+          currentTextSegment += event.content;
+          yield { type: "text_delta", content: event.content };
+          break;
+        case "thinking_delta":
+          if (firstTokenTime === null) {
+            firstTokenTime = performance.now();
+          }
+          flushText();
+          thinkingContent += event.content;
+          currentThinkingSegment += event.content;
+          yield { type: "thinking_delta", content: event.content };
+          break;
+        case "tool_call_delta":
+          yield {
+            type: "tool_call_delta",
+            index: event.index,
+            ...event.id !== void 0 ? { id: event.id } : {},
+            ...event.name !== void 0 ? { name: event.name } : {},
+            ...event.arguments !== void 0 ? { arguments: event.arguments } : {}
+          };
+          break;
+        case "done":
+          flushThinking();
+          flushText();
+          response = event.response;
+          yield { type: "done", response: event.response };
+          break;
+        case "error":
+          yield { type: "error", error: event.error };
+          return null;
+      }
+    }
+  } catch (error) {
+    yield { type: "error", error: error instanceof Error ? error.message : "Unknown error" };
+    return null;
+  }
+  if (!response) {
+    return null;
+  }
+  const toolCalls = response.toolCalls ?? [];
+  for (const tc of toolCalls) {
+    segments.push({ type: "tool_call", toolCallId: tc.id });
+  }
+  const endTime = performance.now();
+  const ttft = ((firstTokenTime ?? endTime) - startTime) / 1e3;
+  const completionTime = (endTime - (firstTokenTime ?? startTime)) / 1e3;
+  const { promptTokens, completionTokens } = response.usage;
+  return {
+    content,
+    thinkingContent,
+    toolCalls,
+    response,
+    segments,
+    timing: {
+      ttft,
+      completionTime,
+      tps: completionTime > 0 ? completionTokens / completionTime : 0,
+      prefillTps: ttft > 0 ? promptTokens / ttft : 0
+    }
+  };
+}
+
+export {
+  streamWithSegments
+};
+//# sourceMappingURL=chunk-J2GP3J3X.js.map

package/dist/{chunk-MDRNKI7D.js → chunk-K454WU7A.js}

@@ -1,21 +1,3 @@
-import {
-  getProject
-} from "./chunk-O4TED6AJ.js";
-import {
-  COMPACTION_PROMPT,
-  TurnMetrics,
-  buildBasePrompt,
-  buildSubAgentSystemPrompt,
-  buildTopLevelSystemPrompt,
-  computeEffectiveTools,
-  consumeStreamGenerator,
-  createChatDoneEvent,
-  createMessageDoneEvent,
-  createMessageStartEvent,
-  createToolCallEvent,
-  createToolResultEvent,
-  streamLLMPure
-} from "./chunk-YUHODMKY.js";
 import {
   startInspectProxy
 } from "./chunk-DL6ZILAF.js";
@@ -35,21 +17,6 @@ import {
 import {
   createMcpTools
 } from "./chunk-NWO6GRYE.js";
-import {
-  getCurrentContextWindowId,
-  getCurrentWindowMessageOptions,
-  getEventStore,
-  getRuntimeConfig
-} from "./chunk-YBWY4DKY.js";
-import {
-  buildContextMessagesFromEventHistory,
-  foldContextState,
-  handleMessageDelta,
-  handleMessageThinking,
-  handleToolCall,
-  handleToolResult,
-  stripOrphanedToolCalls
-} from "./chunk-6PLAWCHQ.js";
 import {
   createChatDoneMessage,
   createChatMessageMessage,
@@ -66,6 +33,42 @@ import {
   deleteSetting,
   getSetting
 } from "./chunk-RFNEDBVO.js";
+import {
+  getProject
+} from "./chunk-O4TED6AJ.js";
+import {
+  COMPACTION_PROMPT,
+  TurnMetrics,
+  buildBasePrompt,
+  buildSubAgentSystemPrompt,
+  buildTopLevelSystemPrompt,
+  computeEffectiveTools,
+  consumeStreamGenerator,
+  createChatDoneEvent,
+  createMessageDoneEvent,
+  createMessageStartEvent,
+  createToolCallEvent,
+  createToolResultEvent,
+  streamLLMPure
+} from "./chunk-YGSBVKFU.js";
+import {
+  getCurrentContextWindowId,
+  getCurrentWindowMessageOptions,
+  getEventStore,
+  getRuntimeConfig
+} from "./chunk-YBWY4DKY.js";
+import {
+  buildContextMessagesFromEventHistory,
+  foldContextState,
+  handleMessageDelta,
+  handleMessageThinking,
+  handleToolCall,
+  handleToolResult,
+  stripOrphanedToolCalls
+} from "./chunk-6PLAWCHQ.js";
+import {
+  logger
+} from "./chunk-K44MW7JJ.js";
 import {
   loadGlobalConfig,
   saveGlobalConfig
@@ -75,10 +78,7 @@ import {
 } from "./chunk-CQGTEGKL.js";
 import {
   modelSupportsVision
-} from "./chunk-UUFEE7VR.js";
-import {
-  logger
-} from "./chunk-K44MW7JJ.js";
+} from "./chunk-V4IE7HJY.js";
 
 // src/server/tools/read.ts
 import { readFile as readFile2, stat, readdir } from "fs/promises";
@@ -205,6 +205,9 @@ async function isPathWithinSandbox(path, workdir, sessionId) {
   }
   return { allowed: false, resolvedPath };
 }
+function looksLikeRegex(str) {
+  return /[*?+[\]\\]/.test(str);
+}
 function extractAbsolutePathsFromCommand(command) {
   if (!command.trim()) {
     return [];
@@ -239,6 +242,9 @@ function extractAbsolutePathsFromCommand(command) {
     if (content.startsWith("/") && content.endsWith("/")) {
       continue;
     }
+    if (content.startsWith("/") && looksLikeRegex(content)) {
+      continue;
+    }
     if (content.startsWith("/")) {
       const resolved = normalize(content);
       if (!isSafePath(resolved)) {
@@ -259,7 +265,7 @@ function extractAbsolutePathsFromCommand(command) {
     if (pathCandidate.includes("__URL__") || pathCandidate.includes("__FILEURL__")) {
       continue;
     }
-    if (pathCandidate.endsWith("/") && pathCandidate.split("/").length <= 2) {
+    if (looksLikeRegex(pathCandidate)) {
       continue;
     }
     const resolved = normalize(pathCandidate);
@@ -2482,8 +2488,15 @@ ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
         ...config.subAgentMetadata ? { subAgentId: config.subAgentMetadata.subAgentId, subAgentType: config.subAgentMetadata.subAgentType } : {}
       })
     );
-    const previousContextTokens = sessionManager.getContextState(sessionId).currentTokens;
-    const modelSettings = currentMaxTokensOverride !== void 0 ? { ...sessionManager.getCurrentModelSettings(), maxTokens: currentMaxTokensOverride } : sessionManager.getCurrentModelSettings();
+    const contextState = sessionManager.getContextState(sessionId);
+    const previousContextTokens = contextState.currentTokens;
+    const contextWindow = sessionManager.getCurrentModelContext();
+    const availableForOutput = Math.max(256, contextWindow - contextState.currentTokens);
+    let modelSettings = currentMaxTokensOverride !== void 0 ? { ...sessionManager.getCurrentModelSettings(), maxTokens: currentMaxTokensOverride } : sessionManager.getCurrentModelSettings();
+    if (modelSettings) {
+      const requestedMaxTokens = modelSettings.maxTokens ?? 16384;
+      modelSettings = { ...modelSettings, maxTokens: Math.min(requestedMaxTokens, availableForOutput) };
+    }
     const streamGen = streamLLMPure({
       messageId: assistantMsgId,
       systemPrompt: assembledRequest.systemPrompt,
@@ -2549,9 +2562,9 @@ ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
     );
     sessionManager.setCurrentContextSize(sessionId, result.usage.promptTokens);
     if (!compacting) {
-      const contextState = sessionManager.getContextState(sessionId);
-      const { shouldCompact, appendCompactionPrompt } = await import("./compactor-SEZEZSML.js");
-      if (shouldCompact(contextState.currentTokens, contextState.maxTokens, runtimeConfig.context.compactionThreshold)) {
+      const contextState2 = sessionManager.getContextState(sessionId);
+      const { shouldCompact, appendCompactionPrompt } = await import("./compactor-JMGSZ4DQ.js");
+      if (shouldCompact(contextState2.currentTokens, contextState2.maxTokens, runtimeConfig.context.compactionThreshold)) {
         appendCompactionPrompt(sessionId, append);
         compacting = true;
         continue;
@@ -2562,8 +2575,8 @@ ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
         truncationRetryCount += 1;
         const currentMaxTokens = result.modelParams?.maxTokens ?? 16384;
         const promptTokens = result.usage.promptTokens;
-        const contextWindow = sessionManager.getCurrentModelContext();
-        const newMaxTokens = Math.min(Math.floor(currentMaxTokens * 1.5), contextWindow - promptTokens - 2048);
+        const contextWindow2 = sessionManager.getCurrentModelContext();
+        const newMaxTokens = Math.min(Math.floor(currentMaxTokens * 1.5), contextWindow2 - promptTokens - 2048);
         currentMaxTokensOverride = newMaxTokens;
         const interimStats = turnMetrics.buildStats(statsIdentity, mode);
         append(
@@ -3355,7 +3368,7 @@ var callSubAgentTool = {
       };
     }
     try {
-      const { getToolRegistryForAgent: getToolRegistryForAgent2 } = await import("./tools-7CKTYL2G.js");
+      const { getToolRegistryForAgent: getToolRegistryForAgent2 } = await import("./tools-4KGLCQJL.js");
       const toolRegistry = getToolRegistryForAgent2(agentDef);
       const turnMetrics = new TurnMetrics();
       const result = await executeSubAgent({
@@ -4161,7 +4174,7 @@ async function computeContextHash(sessionManager, sessionId) {
   const runtimeConfig = getRuntimeConfig();
   const configDir = getGlobalConfigDir(runtimeConfig.mode ?? "production");
   const skills = await getEnabledSkillMetadata(configDir, runtimeConfig.workdir);
-  const { createToolRegistry: createToolRegistry2 } = await import("./tools-7CKTYL2G.js");
+  const { createToolRegistry: createToolRegistry2 } = await import("./tools-4KGLCQJL.js");
   const allTools = createToolRegistry2().definitions;
   const toolFingerprint = getToolFingerprint(allTools);
   const hash = computeDynamicContextHash(instructionContent, skills, toolFingerprint);
@@ -4274,7 +4287,7 @@ var mcpConfigTool = createTool(
       await saveGlobalConfig(mcpConfigMode, { ...globalConfig, mcpServers: updated });
     }
     async function rebuildTools() {
-      const { setMcpTools: setMcpTools2 } = await import("./tools-7CKTYL2G.js");
+      const { setMcpTools: setMcpTools2 } = await import("./tools-4KGLCQJL.js");
       const mcpTools = createMcpTools(mcpManagerForTools);
       setMcpTools2(mcpTools);
     }
@@ -4678,4 +4691,4 @@ export {
   getToolRegistryForAgent,
   createToolRegistry
 };
-//# sourceMappingURL=chunk-MDRNKI7D.js.map
+//# sourceMappingURL=chunk-K454WU7A.js.map

package/dist/chunk-M3RB4IF6.js

@@ -0,0 +1,114 @@
+import {
+  buildModelsUrl,
+  stripVersionPrefix
+} from "./chunk-HNCM3D7Y.js";
+import {
+  logger
+} from "./chunk-K44MW7JJ.js";
+
+// src/server/llm/models.ts
+var modelCache = /* @__PURE__ */ new Map();
+var llmStatus = "unknown";
+var lastActiveUrl = null;
+var CACHE_TTL_MS = 3e4;
+function getCacheKey(url) {
+  return stripVersionPrefix(url);
+}
+async function detectModel(llmBaseUrl, retries = 3, silent = false) {
+  const cacheKey = getCacheKey(llmBaseUrl);
+  const now = Date.now();
+  const cached = modelCache.get(cacheKey);
+  if (cached && now - cached.timestamp < CACHE_TTL_MS) {
+    lastActiveUrl = cacheKey;
+    llmStatus = "connected";
+    return cached.model;
+  }
+  const url = buildModelsUrl(llmBaseUrl);
+  for (let attempt = 1; attempt <= retries; attempt++) {
+    try {
+      if (silent) {
+        logger.debug("Fetching models from LLM server", { url, attempt });
+      }
+      const response = await fetch(url, {
+        signal: AbortSignal.timeout(1e4)
+      });
+      if (!response.ok) {
+        if (silent) {
+          logger.debug("Failed to fetch models from LLM server", { status: response.status, attempt });
+        } else {
+          logger.warn("Failed to fetch models from LLM server", { status: response.status, attempt });
+        }
+        if (attempt < retries) {
+          await new Promise((r) => setTimeout(r, 1e3 * attempt));
+          continue;
+        }
+        llmStatus = "disconnected";
+        return cached?.model ?? null;
+      }
+      const data = await response.json();
+      if (data.data && data.data.length > 0) {
+        const modelData = data.data[0];
+        const modelId = modelData.id;
+        modelCache.set(cacheKey, {
+          model: modelId,
+          modelInfo: modelData,
+          timestamp: now
+        });
+        lastActiveUrl = cacheKey;
+        llmStatus = "connected";
+        if (silent) {
+          logger.debug("Detected LLM model", {
+            model: modelId,
+            maxLen: modelData.max_model_len,
+            root: modelData.root
+          });
+        } else {
+          logger.info("Detected LLM model", {
+            model: modelId,
+            maxLen: modelData.max_model_len,
+            root: modelData.root
+          });
+        }
+        return modelId;
+      }
+      if (silent) {
+        logger.debug("LLM server returned empty models list");
+      } else {
+        logger.warn("LLM server returned empty models list");
+      }
+      llmStatus = "disconnected";
+      return null;
+    } catch (error) {
+      const errMsg = error instanceof Error ? error.message : String(error);
+      if (silent) {
+        logger.debug("Could not detect model from LLM server", { error: errMsg, attempt });
+      } else {
+        logger.warn("Could not detect model from LLM server", { error: errMsg, attempt });
+      }
+      if (attempt < retries) {
+        await new Promise((r) => setTimeout(r, 1e3 * attempt));
+        continue;
+      }
+    }
+  }
+  llmStatus = "disconnected";
+  return cached?.model ?? null;
+}
+function getLlmStatus() {
+  return llmStatus;
+}
+function clearModelCache(url) {
+  if (url) {
+    modelCache.delete(getCacheKey(url));
+  } else {
+    modelCache.clear();
+  }
+  llmStatus = "unknown";
+}
+
+export {
+  detectModel,
+  getLlmStatus,
+  clearModelCache
+};
+//# sourceMappingURL=chunk-M3RB4IF6.js.map