Merge pull request #9467 from continuedev/dallin/context-length-exceeded

fix(cli): context length tweaks and terminal output truncation

Author: RomneyDa

docs/cli/configuration.mdx

@@ -0,0 +1,29 @@
+---
+title: "CLI Configuration"
+description: "Configure Continue CLI behavior with environment variables"
+sidebarTitle: "Configuration"
+---
+
+Continue CLI tools automatically truncate large outputs to prevent excessive token usage. You can customize these limits using environment variables.
+
+## Environment Variables
+
+| Environment Variable | Tool | Default |
+|---------------------|------|--------:|
+| `CONTINUE_CLI_BASH_MAX_OUTPUT_CHARS` | Bash | 50,000 |
+| `CONTINUE_CLI_BASH_MAX_OUTPUT_LINES` | Bash | 1,000 |
+| `CONTINUE_CLI_READ_FILE_MAX_OUTPUT_CHARS` | Read | 100,000 |
+| `CONTINUE_CLI_READ_FILE_MAX_OUTPUT_LINES` | Read | 5,000 |
+| `CONTINUE_CLI_FETCH_MAX_OUTPUT_LENGTH` | Fetch | 20,000 |
+| `CONTINUE_CLI_DIFF_MAX_OUTPUT_LENGTH` | Diff | 50,000 |
+| `CONTINUE_CLI_SEARCH_CODE_MAX_RESULTS` | Search | 100 |
+| `CONTINUE_CLI_SEARCH_CODE_MAX_RESULT_CHARS` | Search | 1,000 |
+
+## Example
+
+```bash
+# Increase limits for verbose build output
+export CONTINUE_CLI_BASH_MAX_OUTPUT_CHARS=100000
+export CONTINUE_CLI_BASH_MAX_OUTPUT_LINES=2000
+cn
+```

docs/docs.json

@@ -91,7 +91,8 @@
               "cli/overview",
               "cli/install",
               "cli/quick-start",
-              "cli/guides"
+              "cli/guides",
+              "cli/configuration"
             ]
           },
           {

extensions/cli/src/commands/chat.ts

@@ -363,7 +363,14 @@ async function processMessage(
   telemetryService.logUserPrompt(userInput.length, userInput);
 
   // Check if auto-compacting is needed BEFORE adding user message
-  if (shouldAutoCompact(services.chatHistory.getHistory(), model)) {
+  // Note: This is a preliminary check without tools/systemMessage context.
+  // The streaming path performs a more accurate check with full context.
+  if (
+    shouldAutoCompact({
+      chatHistory: services.chatHistory.getHistory(),
+      model,
+    })
+  ) {
     const newIndex = await handleAutoCompaction(
       chatHistory,
       model,

extensions/cli/src/stream/streamChatResponse.autoCompaction.test.ts

@@ -99,7 +99,12 @@ describe("handleAutoCompaction", () => {
       wasCompacted: false,
     });
 
-    expect(shouldAutoCompact).toHaveBeenCalledWith(mockChatHistory, mockModel);
+    expect(shouldAutoCompact).toHaveBeenCalledWith({
+      chatHistory: mockChatHistory,
+      model: mockModel,
+      systemMessage: undefined,
+      tools: undefined,
+    });
   });
 
   it("should perform auto-compaction when context limit is approaching", async () => {
@@ -144,7 +149,12 @@ describe("handleAutoCompaction", () => {
       },
     );
 
-    expect(shouldAutoCompact).toHaveBeenCalledWith(mockChatHistory, mockModel);
+    expect(shouldAutoCompact).toHaveBeenCalledWith({
+      chatHistory: mockChatHistory,
+      model: mockModel,
+      systemMessage: undefined,
+      tools: undefined,
+    });
     expect(getAutoCompactMessage).toHaveBeenCalledWith(mockModel);
     expect(compactChatHistory).toHaveBeenCalledWith(
       mockChatHistory,

extensions/cli/src/stream/streamChatResponse.autoCompaction.ts

@@ -1,6 +1,7 @@
 import { ModelConfig } from "@continuedev/config-yaml";
 import { BaseLlmApi } from "@continuedev/openai-adapters";
 import type { ChatHistoryItem } from "core/index.js";
+import type { ChatCompletionTool } from "openai/resources/chat/completions.mjs";
 import React from "react";
 
 import { compactChatHistory } from "../compaction.js";
@@ -27,6 +28,7 @@ interface AutoCompactionOptions {
   format?: "json";
   callbacks?: AutoCompactionCallbacks;
   systemMessage?: string;
+  tools?: ChatCompletionTool[];
 }
 
 /**
@@ -133,9 +135,18 @@ export async function handleAutoCompaction(
     isHeadless = false,
     callbacks,
     systemMessage: providedSystemMessage,
+    tools,
   } = options;
 
-  if (!model || !shouldAutoCompact(chatHistory, model)) {
+  if (
+    !model ||
+    !shouldAutoCompact({
+      chatHistory,
+      model,
+      systemMessage: providedSystemMessage,
+      tools,
+    })
+  ) {
     return { chatHistory, compactionIndex: null, wasCompacted: false };
   }
 

extensions/cli/src/stream/streamChatResponse.compactionHelpers.ts

@@ -1,6 +1,7 @@
 import { ModelConfig } from "@continuedev/config-yaml";
 import { BaseLlmApi } from "@continuedev/openai-adapters";
 import type { ChatHistoryItem } from "core/index.js";
+import type { ChatCompletionTool } from "openai/resources/chat/completions.mjs";
 
 import { services } from "../services/index.js";
 import { ToolCall } from "../tools/index.js";
@@ -17,6 +18,7 @@ export interface CompactionHelperOptions {
   isHeadless: boolean;
   callbacks?: StreamCallbacks;
   systemMessage: string;
+  tools?: ChatCompletionTool[];
 }
 
 /**
@@ -26,8 +28,15 @@ export async function handlePreApiCompaction(
   chatHistory: ChatHistoryItem[],
   options: CompactionHelperOptions,
 ): Promise<{ chatHistory: ChatHistoryItem[]; wasCompacted: boolean }> {
-  const { model, llmApi, isCompacting, isHeadless, callbacks, systemMessage } =
-    options;
+  const {
+    model,
+    llmApi,
+    isCompacting,
+    isHeadless,
+    callbacks,
+    systemMessage,
+    tools,
+  } = options;
 
   if (isCompacting) {
     return { chatHistory, wasCompacted: false };
@@ -41,6 +50,7 @@ export async function handlePreApiCompaction(
         onContent: callbacks?.onContent,
       },
       systemMessage,
+      tools,
     });
 
   if (wasCompacted) {
@@ -67,7 +77,8 @@ export async function handlePostToolValidation(
   chatHistory: ChatHistoryItem[],
   options: CompactionHelperOptions,
 ): Promise<{ chatHistory: ChatHistoryItem[]; wasCompacted: boolean }> {
-  const { model, llmApi, isHeadless, callbacks, systemMessage } = options;
+  const { model, llmApi, isHeadless, callbacks, systemMessage, tools } =
+    options;
 
   if (toolCalls.length === 0) {
     return { chatHistory, wasCompacted: false };
@@ -82,21 +93,14 @@ export async function handlePostToolValidation(
     chatHistory = chatHistorySvc.getHistory();
   }
 
-  // Validate with system message to catch tool result overflow
-  const postToolSystemItem: ChatHistoryItem = {
-    message: {
-      role: "system",
-      content: systemMessage,
-    },
-    contextItems: [],
-  };
-
   const SAFETY_BUFFER = 100;
-  const postToolValidation = validateContextLength(
-    [postToolSystemItem, ...chatHistory],
+  const postToolValidation = validateContextLength({
+    chatHistory,
     model,
-    SAFETY_BUFFER,
-  );
+    safetyBuffer: SAFETY_BUFFER,
+    systemMessage,
+    tools,
+  });
 
   // If tool results pushed us over limit, force compaction regardless of threshold
   if (!postToolValidation.isValid) {
@@ -114,6 +118,7 @@ export async function handlePostToolValidation(
           onContent: callbacks?.onContent,
         },
         systemMessage,
+        tools,
       });
 
     if (wasCompacted) {
@@ -127,11 +132,13 @@ export async function handlePostToolValidation(
       }
 
       // Verify compaction brought us under the limit
-      const postCompactionValidation = validateContextLength(
-        [postToolSystemItem, ...chatHistory],
+      const postCompactionValidation = validateContextLength({
+        chatHistory,
         model,
-        SAFETY_BUFFER,
-      );
+        safetyBuffer: SAFETY_BUFFER,
+        systemMessage,
+        tools,
+      });
 
       if (!postCompactionValidation.isValid) {
         logger.error(
@@ -171,7 +178,8 @@ export async function handleNormalAutoCompaction(
   shouldContinue: boolean,
   options: CompactionHelperOptions,
 ): Promise<{ chatHistory: ChatHistoryItem[]; wasCompacted: boolean }> {
-  const { model, llmApi, isHeadless, callbacks, systemMessage } = options;
+  const { model, llmApi, isHeadless, callbacks, systemMessage, tools } =
+    options;
 
   if (!shouldContinue) {
     return { chatHistory, wasCompacted: false };
@@ -193,6 +201,7 @@ export async function handleNormalAutoCompaction(
         onContent: callbacks?.onContent,
       },
       systemMessage,
+      tools,
     });
 
   if (wasCompacted) {

extensions/cli/src/stream/streamChatResponse.ts

@@ -215,25 +215,17 @@ export async function processStreamingResponse(
 
   let chatHistory = options.chatHistory;
 
-  // Create temporary system message item for validation
-  const systemMessageItem: ChatHistoryItem = {
-    message: {
-      role: "system",
-      content: systemMessage,
-    },
-    contextItems: [],
-  };
-
   // Safety buffer to account for tokenization estimation errors
   const SAFETY_BUFFER = 100;
 
-  // Validate context length INCLUDING system message
-  let historyWithSystem = [systemMessageItem, ...chatHistory];
-  let validation = validateContextLength(
-    historyWithSystem,
+  // Validate context length INCLUDING system message and tools
+  let validation = validateContextLength({
+    chatHistory,
     model,
-    SAFETY_BUFFER,
-  );
+    safetyBuffer: SAFETY_BUFFER,
+    systemMessage,
+    tools,
+  });
 
   // Prune last messages until valid (excluding system message)
   while (chatHistory.length > 1 && !validation.isValid) {
@@ -243,9 +235,14 @@ export async function processStreamingResponse(
     }
     chatHistory = prunedChatHistory;
 
-    // Re-validate with system message
-    historyWithSystem = [systemMessageItem, ...chatHistory];
-    validation = validateContextLength(historyWithSystem, model, SAFETY_BUFFER);
+    // Re-validate with system message and tools
+    validation = validateContextLength({
+      chatHistory,
+      model,
+      safetyBuffer: SAFETY_BUFFER,
+      systemMessage,
+      tools,
+    });
   }
 
   if (!validation.isValid) {
@@ -464,23 +461,24 @@ export async function streamChatResponse(
       services.toolPermissions.getState().currentMode,
     );
 
-    // Pre-API auto-compaction checkpoint
+    // Recompute tools on each iteration to handle mode changes during streaming
+    const tools = await getRequestTools(isHeadless);
+
+    // Pre-API auto-compaction checkpoint (now includes tools)
     const preCompactionResult = await handlePreApiCompaction(chatHistory, {
       model,
       llmApi,
       isCompacting,
       isHeadless,
       callbacks,
       systemMessage,
+      tools,
     });
     chatHistory = preCompactionResult.chatHistory;
     if (preCompactionResult.wasCompacted) {
       compactionOccurredThisTurn = true;
     }
 
-    // Recompute tools on each iteration to handle mode changes during streaming
-    const tools = await getRequestTools(isHeadless);
-
     logger.debug("Tools prepared", {
       toolCount: tools.length,
       toolNames: tools.map((t) => t.function.name),
@@ -541,6 +539,7 @@ export async function streamChatResponse(
         isHeadless,
         callbacks,
         systemMessage,
+        tools,
       },
     );
     chatHistory = postToolResult.chatHistory;
@@ -559,6 +558,7 @@ export async function streamChatResponse(
         isHeadless,
         callbacks,
         systemMessage,
+        tools,
       },
     );
     chatHistory = compactionResult.chatHistory;

extensions/cli/src/tools/fetch.test.ts

@@ -48,7 +48,7 @@ describe("fetchTool", () => {
     );
   });
 
-  it("should handle truncation warnings from core implementation", async () => {
+  it("should filter out truncation warnings from core implementation", async () => {
     const mockContextItems: ContextItem[] = [
       {
         name: "Long Page",
@@ -68,9 +68,8 @@ describe("fetchTool", () => {
 
     const result = await fetchTool.run({ url: "https://example.com" });
 
-    expect(result).toBe(
-      "This is the main content that was truncated.\n\nThe content from https://example.com was truncated because it exceeded the 20000 character limit.",
-    );
+    // Truncation warnings are filtered out - only the main content is returned
+    expect(result).toBe("This is the main content that was truncated.");
   });
 
   it("should handle multiple content items", async () => {