@linnlabs/linnkit 0.8.0 → 0.10.0

package/dist/context-manager.cjs

@@ -499,6 +499,7 @@ var AgentSpecBudgetPolicy = zod.z.object({
 });
 var AgentSpecToolHistoryPolicy = zod.z.object({
   strategy: zod.z.enum(["per-pair", "per-run", "none"]).optional(),
+  retentionMode: zod.z.enum(["drop", "compress"]).optional(),
   keepLatestToolPairs: zod.z.number().int().nonnegative().optional(),
   keepLatestRuns: zod.z.number().int().nonnegative().optional(),
   maxInteractionGroups: zod.z.number().int().nonnegative().optional(),
@@ -613,12 +614,13 @@ var AgentSpecContextPolicy = zod.z.object({
 var DEFAULT_CONTEXT_POLICY = {
   profileId: "agent",
   budget: {
-    maxTokens: 12e4,
+    maxTokens: 232e3,
     reservedForResponse: 2400,
     workingMemoryBudgetPercentage: 0.7
   },
   toolHistory: {
     strategy: "per-run",
+    retentionMode: "drop",
     keepLatestToolPairs: 2,
     keepLatestRuns: 1,
     maxInteractionGroups: 12,
@@ -1774,6 +1776,157 @@ function findLastIndex(items, predicate) {
   return -1;
 }
 
+// src/context-manager/shared/preprocessors/currentTurnMessageAssembler.ts
+var CurrentTurnMessageAssembler = class extends BasePreprocessor {
+  name = "CurrentTurnMessageAssembler";
+  description = "Assembles current-turn fences into system_prompt and user_input messages";
+  priority = 10;
+  fenceRegistry;
+  constructor(options) {
+    super();
+    this.fenceRegistry = options.fenceRegistry;
+  }
+  async process(messages, _context) {
+    const systemResult = assembleSystemPrompt({ messages, fenceRegistry: this.fenceRegistry });
+    const userResult = assembleCurrentUserInput({
+      messages: systemResult.messages,
+      fenceRegistry: this.fenceRegistry
+    });
+    const modifiedCount = systemResult.modifiedCount + userResult.modifiedCount;
+    return this.createResult(
+      messages,
+      userResult.messages,
+      modifiedCount > 0 ? ["current_turn_message_assembly"] : [],
+      modifiedCount
+    );
+  }
+};
+function assembleSystemPrompt(params) {
+  const systemIndex = params.messages.findIndex(
+    (message) => message.role === "system" && message.type === "system_prompt"
+  );
+  if (systemIndex === -1) {
+    return { messages: [...params.messages], modifiedCount: 0 };
+  }
+  const fences = collectAdjacentFences({
+    messages: params.messages,
+    startIndex: systemIndex + 1,
+    direction: 1,
+    fenceRegistry: params.fenceRegistry,
+    matches: (descriptor) => descriptor.llmRole === "system" && descriptor.placement === "after-system"
+  });
+  if (fences.length === 0) {
+    return { messages: [...params.messages], modifiedCount: 0 };
+  }
+  const indexesToRemove = new Set(fences.map((fence) => fence.index));
+  const systemPrompt = params.messages[systemIndex];
+  const assembledSystemPrompt = {
+    ...systemPrompt,
+    content: [
+      systemPrompt.content,
+      ...fences.map(formatFence)
+    ].join("\n\n"),
+    metadata: {
+      ...systemPrompt.metadata,
+      assembledFenceKinds: fences.map((fence) => fence.message.metadata?.fenceKind).filter((kind) => typeof kind === "string")
+    }
+  };
+  return {
+    messages: params.messages.map(
+      (message, index) => index === systemIndex ? assembledSystemPrompt : message
+    ).filter((_, index) => !indexesToRemove.has(index)),
+    modifiedCount: fences.length + 1
+  };
+}
+function assembleCurrentUserInput(params) {
+  const currentUserIndex = findLastUserInputIndex(params.messages);
+  if (currentUserIndex === -1) {
+    return { messages: [...params.messages], modifiedCount: 0 };
+  }
+  const beforeFences = collectAdjacentFences({
+    messages: params.messages,
+    startIndex: currentUserIndex - 1,
+    direction: -1,
+    fenceRegistry: params.fenceRegistry,
+    matches: (descriptor) => isUserCurrentTurnPlacement(descriptor, "before-current-user")
+  }).reverse();
+  const afterFences = collectAdjacentFences({
+    messages: params.messages,
+    startIndex: currentUserIndex + 1,
+    direction: 1,
+    fenceRegistry: params.fenceRegistry,
+    matches: (descriptor) => isUserCurrentTurnPlacement(descriptor, "after-current-user")
+  });
+  if (beforeFences.length === 0 && afterFences.length === 0) {
+    return { messages: [...params.messages], modifiedCount: 0 };
+  }
+  const indexesToRemove = /* @__PURE__ */ new Set([
+    ...beforeFences.map((fence) => fence.index),
+    ...afterFences.map((fence) => fence.index)
+  ]);
+  const currentUser = params.messages[currentUserIndex];
+  const assembledUser = {
+    ...currentUser,
+    content: [
+      ...beforeFences.map(formatFence),
+      wrapUserRequest(currentUser.content),
+      ...afterFences.map(formatFence)
+    ].join("\n\n"),
+    metadata: {
+      ...currentUser.metadata,
+      assembledFenceKinds: [
+        ...beforeFences.map((fence) => fence.message.metadata?.fenceKind),
+        ...afterFences.map((fence) => fence.message.metadata?.fenceKind)
+      ].filter((kind) => typeof kind === "string")
+    }
+  };
+  return {
+    messages: params.messages.map(
+      (message, index) => index === currentUserIndex ? assembledUser : message
+    ).filter((_, index) => !indexesToRemove.has(index)),
+    modifiedCount: indexesToRemove.size + 1
+  };
+}
+function findLastUserInputIndex(messages) {
+  for (let index = messages.length - 1; index >= 0; index -= 1) {
+    const message = messages[index];
+    if (message.role === "user" && message.type === "user_input") {
+      return index;
+    }
+  }
+  return -1;
+}
+function collectAdjacentFences(params) {
+  const result = [];
+  for (let index = params.startIndex; index >= 0 && index < params.messages.length; index += params.direction) {
+    const message = params.messages[index];
+    if (message.type !== "context_injection") {
+      break;
+    }
+    const fenceKind = message.metadata?.fenceKind;
+    if (!fenceKind) {
+      break;
+    }
+    const descriptor = params.fenceRegistry.get(fenceKind);
+    if (!descriptor || !params.matches(descriptor)) {
+      break;
+    }
+    result.push({ index, message, descriptor });
+  }
+  return result;
+}
+function isUserCurrentTurnPlacement(descriptor, placement) {
+  return descriptor.llmRole === "user" && descriptor.placement === placement;
+}
+function formatFence(fence) {
+  return fence.descriptor.formatter(fence.message.content, fence.message.metadata?.fenceAttrs ?? {});
+}
+function wrapUserRequest(content) {
+  return `<user_request>
+${content.trim()}
+</user_request>`;
+}
+
 // src/context-manager/shared/providers/base.ts
 var TOOL_HISTORY_OVERFLOW_ERROR_CODE = "TOOL_HISTORY_OVERFLOW";
 var SUMMARIZATION_FAILED_ERROR_CODE = "SUMMARIZATION_FAILED";
@@ -2576,7 +2729,7 @@ __export(config_exports, {
 var AGENT_CONTEXT_BUILDER_CONFIG = {
   // === 绝对Token限制设置 ===
   /** 默认最大Token预算上限 */
-  DEFAULT_MAX_TOKENS: 12e4,
+  DEFAULT_MAX_TOKENS: 232e3,
   /** 响应预留Token数 */
   RESERVED_FOR_RESPONSE: 2400,
   // === Agent专用工作记忆填充策略 ===
@@ -3967,7 +4120,7 @@ var AgentWorkingMemoryProvider = class extends BaseContextProvider2 {
     let workingMemoryTokens = coreTokens;
     let processedCount = 0;
     const strategiesApplied = [];
-    const workingMemoryBudget = Math.floor(availableBudget * config.WORKING_MEMORY_BUDGET_PERCENTAGE);
+    const workingMemoryBudget = Math.floor(context.totalBudget * config.WORKING_MEMORY_BUDGET_PERCENTAGE);
     const remainingBudget = workingMemoryBudget - coreTokens;
     if (remainingBudget <= 0 && config.MIN_TOOL_INTERACTIONS_TO_KEEP <= 0) {
       this.debug("\u26A0\uFE0F \u6838\u5FC3\u4E0A\u4E0B\u6587\u5DF2\u7528\u5C3D\u9884\u7B97\uFF0C\u8DF3\u8FC7\u5DE5\u4F5C\u8BB0\u5FC6\u586B\u5145", {
@@ -5110,6 +5263,7 @@ __export(orchestration_exports, {
 var preprocessors_exports = {};
 __export(preprocessors_exports, {
   BasePreprocessor: () => BasePreprocessor2,
+  CurrentTurnMessageAssembler: () => CurrentTurnMessageAssembler,
   FenceLifetimePreprocessor: () => FenceLifetimePreprocessor,
   HistoryPurificationPreprocessor: () => HistoryPurificationPreprocessor,
   PreprocessorPipeline: () => PreprocessorPipeline,
@@ -5161,7 +5315,7 @@ var DEFAULT_KEEP_LATEST_RUNS = 1;
 var DEFAULT_MAX_INTERACTION_GROUPS = 12;
 var ToolHistoryCompressorPreprocessor = class extends BasePreprocessor2 {
   name = "ToolHistoryCompressorPreprocessor";
-  description = "\u5DE5\u5177\u5386\u53F2\u538B\u7F29\u5904\u7406\u5668 - \u5C06\u8F83\u65E9\u7684\u5386\u53F2\u5DE5\u5177\u8C03\u7528\u5BF9\u538B\u7F29\u4E3A\u81EA\u7136\u8BED\u8A00\u8BB0\u5F55\u6D88\u606F";
+  description = "\u5DE5\u5177\u5386\u53F2\u4FDD\u7559\u5904\u7406\u5668 - \u6309\u7B56\u7565\u4FDD\u7559\u3001\u5220\u9664\u6216\u538B\u7F29\u8F83\u65E9\u7684\u5386\u53F2\u5DE5\u5177\u8C03\u7528\u5BF9";
   priority = 0;
   summarizer = createDefaultToolOutputSummarizer();
   options;
@@ -5169,6 +5323,7 @@ var ToolHistoryCompressorPreprocessor = class extends BasePreprocessor2 {
     super();
     this.options = {
       strategy: options.strategy ?? "per-run",
+      retentionMode: options.retentionMode ?? "drop",
       keepLatestToolPairs: normalizeNonNegativeInteger4(
         options.keepLatestToolPairs,
         DEFAULT_KEEP_LATEST_TOOL_PAIRS
@@ -5185,7 +5340,7 @@ var ToolHistoryCompressorPreprocessor = class extends BasePreprocessor2 {
     };
   }
   async process(messages, context) {
-    this.debug("\u{1F527} \u5F00\u59CB\u5DE5\u5177\u5386\u53F2\u538B\u7F29\u5904\u7406", {
+    this.debug("\u{1F527} \u5F00\u59CB\u5DE5\u5177\u5386\u53F2\u4FDD\u7559\u5904\u7406", {
       \u539F\u59CB\u6D88\u606F\u6570: messages.length
     }, context);
     const currentRunStartIndex = findCurrentRunStartIndex(messages);
@@ -5195,17 +5350,18 @@ var ToolHistoryCompressorPreprocessor = class extends BasePreprocessor2 {
       \u5386\u53F2\u6D88\u606F\u6570: historyMessages.length,
       \u5F53\u524D\u8F6E\u6B21\u6D88\u606F\u6570: currentRunMessages.length,
       \u5F53\u524D\u8F6E\u6B21\u8D77\u70B9\u7D22\u5F15: currentRunStartIndex,
-      \u5DE5\u5177\u538B\u7F29\u7B56\u7565: this.options.strategy
+      \u5DE5\u5177\u4FDD\u7559\u7B56\u7565: this.options.strategy,
+      \u5386\u53F2\u5DE5\u5177\u4FDD\u7559\u6A21\u5F0F: this.options.retentionMode
     }, context);
-    const compressedHistory = this.compressToolCallPairsInHistory(historyMessages, context);
-    const finalMessages = [...compressedHistory, ...currentRunMessages];
+    const processedHistory = this.processToolCallPairsInHistory(historyMessages, context);
+    const finalMessages = [...processedHistory, ...currentRunMessages];
     const originalCount = messages.length;
-    const compressedCount = finalMessages.length;
-    const removedCount = originalCount - compressedCount;
-    const appliedStrategies = removedCount > 0 ? ["tool_history_compression"] : [];
-    this.debug("\u2705 \u5DE5\u5177\u5386\u53F2\u538B\u7F29\u5B8C\u6210", {
+    const processedCount = finalMessages.length;
+    const removedCount = originalCount - processedCount;
+    const appliedStrategies = removedCount > 0 ? [this.options.retentionMode === "compress" ? "tool_history_compression" : "tool_history_drop"] : [];
+    this.debug("\u2705 \u5DE5\u5177\u5386\u53F2\u4FDD\u7559\u5904\u7406\u5B8C\u6210", {
       \u539F\u59CB\u6D88\u606F: originalCount,
-      \u538B\u7F29\u540E\u6D88\u606F: compressedCount,
+      \u5904\u7406\u540E\u6D88\u606F: processedCount,
       \u51CF\u5C11\u6D88\u606F: removedCount,
       Token\u8282\u7701\u4F30\u8BA1: `\u7EA6${removedCount * 50}\u4E2AToken`
     }, context);
@@ -5226,7 +5382,7 @@ var ToolHistoryCompressorPreprocessor = class extends BasePreprocessor2 {
     }
     return false;
   }
-  compressToolCallPairsInHistory(historyMessages, context) {
+  processToolCallPairsInHistory(historyMessages, context) {
     if (historyMessages.length < 2) {
       return historyMessages;
     }
@@ -5247,8 +5403,10 @@ var ToolHistoryCompressorPreprocessor = class extends BasePreprocessor2 {
       if (!group.isComplete || keepAnchorIds.has(group.anchorId)) {
         continue;
       }
-      const compressedMessage = this.compressToolInteractionGroup(group, context);
-      replacementMap.set(group.assistantIndex, compressedMessage);
+      if (this.options.retentionMode === "compress") {
+        const compressedMessage = this.compressToolInteractionGroup(group, context);
+        replacementMap.set(group.assistantIndex, compressedMessage);
+      }
       for (const messageIndex of group.messageIndexes) {
         messagesToRemove.add(messageIndex);
       }
@@ -5667,6 +5825,7 @@ function createDefaultAgentPreprocessorRegistry(options = {}) {
   registry.register(new ToolReplayProtocolGuardPreprocessor({ policy: options.providerReplay }));
   registry.register(new HistoryPurificationPreprocessor({ logPrefix: "Agent-HistoryPurification" }));
   if (options.fenceRegistry) {
+    registry.register(new CurrentTurnMessageAssembler({ fenceRegistry: options.fenceRegistry }));
     registry.register(new FenceLifetimePreprocessor({ fenceRegistry: options.fenceRegistry }));
   }
   return registry;
@@ -6782,6 +6941,18 @@ var AgentMessageOrchestrator = class {
       tokenizer: this.options.tokenizer,
       tokenizerModelId: this.resolvePreprocessorModel(request)
     });
+    return contextBuilderConfig;
+  }
+  resolveEffectiveContextBudget(contextBuilderConfig) {
+    const maxTokens = contextBuilderConfig.DEFAULT_MAX_TOKENS ?? this.options.tokenBudget.maxTokens;
+    const reservedForResponse = contextBuilderConfig.RESERVED_FOR_RESPONSE ?? this.options.tokenBudget.reservedForResponse;
+    return {
+      maxTokens,
+      reservedForResponse,
+      // 中文备注：ContextManager 接收的是“可放入上下文的输入预算”，不是模型完整窗口。
+      // 因此单个 agent 通过 contextPolicy.budget 覆盖预算时，这里必须同步使用覆盖后的值。
+      totalBudget: maxTokens - reservedForResponse
+    };
   }
   async processAgentConversation(request, history, toolManager, callbacks, extraOptions) {
     const historyCount = history.length;
@@ -6800,7 +6971,8 @@ var AgentMessageOrchestrator = class {
       const allMessages = this.buildCompleteMessageList(request, historyMessages);
       this.debug("Built complete message list", { totalCount: allMessages.length });
       const contextPolicy = this.resolveContextPolicy(request);
-      this.applyContextPolicy(request, contextPolicy);
+      const contextBuilderConfig = this.applyContextPolicy(request, contextPolicy);
+      const effectiveContextBudget = this.resolveEffectiveContextBudget(contextBuilderConfig);
       const preprocessorPipeline = this.buildPreprocessorPipelineForRequest(toolManager, request, contextPolicy);
       const modelId = this.resolvePreprocessorModel(request);
       preprocessorPipeline.updateContext({
@@ -6828,7 +7000,8 @@ var AgentMessageOrchestrator = class {
         callbacks,
         void 0,
         extraOptions?.generate,
-        contextPolicy
+        contextPolicy,
+        effectiveContextBudget.totalBudget
       );
       this.debug("Context built", { afterContextCount: contextResult.messages.length });
       if (this.options.processing.debugMode) {
@@ -6861,7 +7034,7 @@ var AgentMessageOrchestrator = class {
           processedCount: contextResult.messages.length,
           tokenUsage: {
             estimated: contextResult.tokenUsage.used,
-            budget: this.options.tokenBudget.maxTokens,
+            budget: effectiveContextBudget.totalBudget,
             remaining: contextResult.tokenUsage.remaining
           },
           processingStats: contextResult.processingStats,
@@ -6881,13 +7054,13 @@ var AgentMessageOrchestrator = class {
   async runPreprocessorPipeline(preprocessorPipeline, messages) {
     return preprocessorPipeline.process(messages);
   }
-  async buildContextFromPreprocessedMessages(request, conversationSession, messages, callbacks, phaseOverride, generate, contextPolicy) {
-    const totalBudget = this.options.tokenBudget.maxTokens - this.options.tokenBudget.reservedForResponse;
+  async buildContextFromPreprocessedMessages(request, conversationSession, messages, callbacks, phaseOverride, generate, contextPolicy, totalBudget) {
+    const resolvedTotalBudget = totalBudget ?? this.options.tokenBudget.maxTokens - this.options.tokenBudget.reservedForResponse;
     const contextResult = await this.agentContextManager.buildContextFromPreprocessedMessages(
       request,
       conversationSession,
       messages,
-      totalBudget,
+      resolvedTotalBudget,
       callbacks,
       phaseOverride,
       generate,
@@ -6968,22 +7141,33 @@ var BaseAgentTask = class {
       });
     }
     messages.push(...fenceMessages["after-system"]);
+    const currentUserIndex = findCurrentUserInputIndex(history, request.query);
+    const historyBeforeCurrentUser = currentUserIndex === -1 ? history : history.slice(0, currentUserIndex);
+    const currentUserMessage = currentUserIndex === -1 ? this.createCurrentUserMessage(request.query) : history[currentUserIndex];
+    const historyAfterCurrentUser = currentUserIndex === -1 ? [] : history.slice(currentUserIndex + 1);
+    messages.push(...historyBeforeCurrentUser);
     messages.push(...fenceMessages["before-current-user"]);
-    const hasUserMessageInHistory = history.some((m) => m.role === "user");
-    if (!hasUserMessageInHistory && request.query && request.query.trim()) {
-      messages.push({
-        id: generateMessageId(),
-        role: "user",
-        type: "user_input",
-        content: request.query.trim(),
-        timestamp: Date.now()
-      });
+    if (currentUserMessage) {
+      messages.push(currentUserMessage);
     }
     messages.push(...fenceMessages["after-current-user"]);
-    messages.push(...history);
+    messages.push(...historyAfterCurrentUser);
     this.insertAfterLastToolResult(messages, fenceMessages["after-last-tool-result"]);
     return messages;
   }
+  createCurrentUserMessage(query) {
+    const content = query.trim();
+    if (!content) {
+      return null;
+    }
+    return {
+      id: generateMessageId(),
+      role: "user",
+      type: "user_input",
+      content,
+      timestamp: Date.now()
+    };
+  }
   createFenceMessages(fences) {
     const grouped = {
       "after-system": [],
@@ -7048,6 +7232,19 @@ function findLastIndex2(items, predicate) {
   }
   return -1;
 }
+function findCurrentUserInputIndex(history, query) {
+  const normalizedQuery = query.trim();
+  if (!normalizedQuery) {
+    return -1;
+  }
+  for (let index = history.length - 1; index >= 0; index -= 1) {
+    const message = history[index];
+    if (message.role === "user" && message.type === "user_input" && message.content.trim() === normalizedQuery) {
+      return index;
+    }
+  }
+  return -1;
+}
 
 // src/context-manager/profiles/agent/tools/index.ts
 var tools_exports2 = {};
@@ -8661,6 +8858,7 @@ exports.ChatWorkingMemoryProvider = WorkingMemoryProvider;
 exports.ContextProviderError = ContextProviderError;
 exports.ContextProviderRegistry = ContextProviderRegistry;
 exports.ContextTraceCollector = ContextTraceCollector;
+exports.CurrentTurnMessageAssembler = CurrentTurnMessageAssembler;
 exports.DEFAULT_MUST_KEEP_POLICY = DEFAULT_MUST_KEEP_POLICY;
 exports.FenceLifetimePreprocessor = FenceLifetimePreprocessor;
 exports.HistoryPurificationPreprocessor = HistoryPurificationPreprocessor;