diff --git a/extensions/cli/src/stream/streamChatResponse.ts b/extensions/cli/src/stream/streamChatResponse.ts
index 2c7ef956611..f55e6713f47 100644
--- a/extensions/cli/src/stream/streamChatResponse.ts
+++ b/extensions/cli/src/stream/streamChatResponse.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 { compileChatMessages } from "core/llm/countTokens.js";
 import { convertFromUnifiedHistoryWithSystemMessage } from "core/util/messageConversion.js";
 import * as dotenv from "dotenv";
 import type {
@@ -18,7 +19,6 @@ import {
   withExponentialBackoff,
 } from "../util/exponentialBackoff.js";
 import { logger } from "../util/logger.js";
-import { validateContextLength } from "../util/tokenizer.js";
 
 import { getAllTools, handleToolCalls } from "./handleToolCalls.js";
 import { handleAutoCompaction } from "./streamChatResponse.autoCompaction.js";
@@ -148,20 +148,57 @@ export async function processStreamingResponse(
     tools,
   } = options;
 
-  // Validate context length before making the request
-  const validation = validateContextLength(chatHistory, model);
-  if (!validation.isValid) {
-    throw new Error(`Context length validation failed: ${validation.error}`);
-  }
-
-  // Get fresh system message and inject it
+  // Get fresh system message
   const systemMessage = await services.systemMessage.getSystemMessage(
     services.toolPermissions.getState().currentMode,
   );
-  const openaiChatHistory = convertFromUnifiedHistoryWithSystemMessage(
+
+  // Convert unified history to ChatMessage format for compileChatMessages
+  const openaiMessages = convertFromUnifiedHistoryWithSystemMessage(
     chatHistory,
     systemMessage,
   ) as ChatCompletionMessageParam[];
+
+  // Convert to ChatMessage format and use compileChatMessages to handle pruning
+  const contextLength = model.contextLength || 4096;
+  const maxTokens = model.defaultCompletionOptions?.maxTokens || 1024;
+
+  const chatMessages = openaiMessages.map((msg) => ({
+    role: msg.role,
+    content: msg.content || "",
+    ...("tool_calls" in msg && msg.tool_calls
+      ? { toolCalls: msg.tool_calls }
+      : {}),
+    ...("tool_call_id" in msg && msg.tool_call_id
+      ? { toolCallId: msg.tool_call_id }
+      : {}),
+  }));
+
+  const result = compileChatMessages({
+    modelName: model.model,
+    msgs: chatMessages,
+    knownContextLength: contextLength,
+    maxTokens,
+    supportsImages: false,
+    tools,
+  });
+
+  if (result.didPrune) {
+    logger.info("Chat history pruned to fit context length", {
+      originalLength: chatHistory.length,
+      prunedLength: result.compiledChatMessages.length,
+      contextPercentage: `${(result.contextPercentage * 100).toFixed(1)}%`,
+    });
+  }
+
+  // Convert back to OpenAI format
+  const openaiChatHistory = result.compiledChatMessages.map((msg: any) => ({
+    role: msg.role,
+    content: msg.content,
+    ...(msg.toolCalls ? { tool_calls: msg.toolCalls } : {}),
+    ...(msg.toolCallId ? { tool_call_id: msg.toolCallId } : {}),
+  })) as ChatCompletionMessageParam[];
+
   const requestStartTime = Date.now();
 
   const streamFactory = async (retryAbortSignal: AbortSignal) => {
diff --git a/extensions/cli/src/tools/runTerminalCommand.ts b/extensions/cli/src/tools/runTerminalCommand.ts
index 00204d55bd1..7f9d2336881 100644
--- a/extensions/cli/src/tools/runTerminalCommand.ts
+++ b/extensions/cli/src/tools/runTerminalCommand.ts
@@ -13,6 +13,32 @@ import {
 
 import { Tool } from "./types.js";
 
+// Maximum number of lines and characters to return from command output
+const MAX_OUTPUT_LINES = 5000;
+const MAX_OUTPUT_CHARS = 200000;
+
+// Helper function to truncate command output by both lines and characters
+// Keeps the LAST X lines/chars to capture test/install outcomes
+function truncateOutput(output: string): string {
+  const lines = output.split("\n");
+  let truncated = output;
+  let truncationMsg = "";
+
+  // First check character limit - keep last X characters
+  if (output.length > MAX_OUTPUT_CHARS) {
+    const startIndex = output.length - MAX_OUTPUT_CHARS;
+    truncated = output.substring(startIndex);
+    truncationMsg = `[Output truncated: showing last ${MAX_OUTPUT_CHARS} characters of ${output.length} total]\n\n`;
+  }
+  // Then check line limit (only if not already truncated by characters) - keep last X lines
+  else if (lines.length > MAX_OUTPUT_LINES) {
+    truncated = lines.slice(-MAX_OUTPUT_LINES).join("\n");
+    truncationMsg = `[Output truncated: showing last ${MAX_OUTPUT_LINES} lines of ${lines.length} total]\n\n`;
+  }
+
+  return truncationMsg ? truncationMsg + truncated : truncated;
+}
+
 // Helper function to use login shell on Unix/macOS and PowerShell on Windows
 function getShellCommand(command: string): { shell: string; args: string[] } {
   if (process.platform === "win32") {
@@ -99,18 +125,9 @@ Commands are automatically executed from the current working directory (${proces
           let output = stdout + (stderr ? `\nStderr: ${stderr}` : "");
           output += `\n\n[Command timed out after ${TIMEOUT_MS / 1000} seconds of no output]`;
 
-          // Truncate output if it has too many lines
-          const lines = output.split("\n");
-          if (lines.length > 5000) {
-            const truncatedOutput = lines.slice(0, 5000).join("\n");
-            resolve(
-              truncatedOutput +
-                `\n\n[Output truncated to first 5000 lines of ${lines.length} total]`,
-            );
-            return;
-          }
-
-          resolve(output);
+          // Truncate output by both lines and characters
+          const truncatedOutput = truncateOutput(output);
+          resolve(truncatedOutput);
         }, TIMEOUT_MS);
       };
 
@@ -155,18 +172,9 @@ Commands are automatically executed from the current working directory (${proces
           output = stdout + `\nStderr: ${stderr}`;
         }
 
-        // Truncate output if it has too many lines
-        const lines = output.split("\n");
-        if (lines.length > 5000) {
-          const truncatedOutput = lines.slice(0, 5000).join("\n");
-          resolve(
-            truncatedOutput +
-              `\n\n[Output truncated to first 5000 lines of ${lines.length} total]`,
-          );
-          return;
-        }
-
-        resolve(output);
+        // Truncate output by both lines and characters
+        const truncatedOutput = truncateOutput(output);
+        resolve(truncatedOutput);
       });
 
       child.on("error", (error) => {