fix(anthropic): Work around Anthropic 500 bug

Anthropic has a bug where if we pass them back one of their bash_code_execution_output blocks which contains a file output, they return a 500. I have reported the error to them, but in the meantime this works around the problem by dropping the problematic content blocks before we send them to anthropic

Also adds an integration test that exercises this PR as well as:

- langchain-ai#9108
- langchain-ai#9109

Author: pokey

libs/providers/langchain-anthropic/src/tests/chat_models-code_execution.int.test.ts

@@ -1,12 +1,36 @@
-import { describe, it, expect } from "vitest";
+import { describe, it, expect, beforeEach } from "vitest";
 import Anthropic from "@anthropic-ai/sdk";
 import { HumanMessage, AIMessage } from "@langchain/core/messages";
 import { ChatAnthropic } from "../chat_models.js";
+import { extractGeneratedFiles } from "../utils/extractGeneratedFiles.js";
 import fs from "fs";
 import path from "path";
 import os from "os";
 
 describe("Files API with Code Execution", () => {
+  let model: ReturnType<ChatAnthropic["bindTools"]>;
+
+  beforeEach(() => {
+    // Create ChatAnthropic model with code execution beta header and bind tools
+    const baseModel = new ChatAnthropic({
+      model: "claude-3-5-haiku-20241022",
+      temperature: 0,
+      clientOptions: {
+        defaultHeaders: {
+          "anthropic-beta": "code-execution-2025-08-25,files-api-2025-04-14",
+        },
+      },
+    });
+
+    // Bind the code_execution tool
+    model = baseModel.bindTools([
+      {
+        type: "code_execution_20250825" as const,
+        name: "code_execution",
+      },
+    ]);
+  });
+
   it("should handle createMessageWithFiles using container_upload blocks", async () => {
     // Create Anthropic client for Files API
     const anthropicClient = new Anthropic({
@@ -26,23 +50,6 @@ describe("Files API with Code Execution", () => {
         file: fs.createReadStream(tmpFilePath),
       });
 
-      // Create ChatAnthropic model with code execution beta header
-      const model = new ChatAnthropic({
-        model: "claude-3-5-haiku-20241022",
-        temperature: 0,
-        clientOptions: {
-          defaultHeaders: {
-            "anthropic-beta": "code-execution-2025-08-25,files-api-2025-04-14",
-          },
-        },
-      });
-
-      // Define the built-in code_execution tool
-      const codeExecutionTool = {
-        type: "code_execution_20250825" as const,
-        name: "code_execution",
-      };
-
       // Create a message with container_upload block
       const message = new HumanMessage({
         content: [
@@ -55,9 +62,7 @@ describe("Files API with Code Execution", () => {
       });
 
       // Invoke the model with the file
-      const result = await model.invoke([message], {
-        tools: [codeExecutionTool],
-      });
+      const result = await model.invoke([message]);
 
       // Verify the result is an AIMessage
       expect(result).toBeInstanceOf(AIMessage);
@@ -86,26 +91,64 @@ describe("Files API with Code Execution", () => {
         )
       ).toBe(true);
 
-      // The response should contain text with the calculated average
-      const textBlocks = content.filter(
-        (block) =>
-          typeof block === "object" && "type" in block && block.type === "text"
-      );
-      const responseText = textBlocks
-        .map((block) =>
-          typeof block === "object" && "text" in block ? block.text : ""
-        )
-        .join(" ")
-        .toLowerCase();
-
       // The response should mention the average age (30)
-      expect(responseText).toMatch(/average|mean/i);
-      expect(responseText).toMatch(/30/);
+      expect(result.text).toMatch(/average|mean/i);
+      expect(result.text).toMatch(/30/);
     } finally {
       // Clean up temporary file
       if (fs.existsSync(tmpFilePath)) {
         fs.unlinkSync(tmpFilePath);
       }
     }
   }, 60000);
+
+  it("should pass container and file outputs across multiple turns", async () => {
+    // First invocation: Calculate mean and avg, store to file
+    const firstMessage = new HumanMessage(
+      "Calculate the mean and average of these numbers: 10, 20, 30, 40, 50. Store the results in a file called 'results.txt'."
+    );
+
+    const firstResult = await model.invoke([firstMessage]);
+
+    // Verify first result succeeded
+    expect(firstResult).toBeInstanceOf(AIMessage);
+
+    // Extract container ID from first response
+    const container = (firstResult as AIMessage).additional_kwargs
+      ?.container as { id: string; expires_at: string } | undefined;
+    expect(container?.id).toBeTruthy();
+
+    // Verify file output was created using extractGeneratedFilesAnthropic
+    const fileIds = extractGeneratedFiles(
+      firstResult as unknown as Anthropic.Beta.BetaMessage
+    );
+    expect(fileIds.length).toBeGreaterThan(0);
+
+    // Second invocation: Read the file with same container
+    // This should succeed because we apply the workaround in message_inputs.ts
+    const secondMessage = new HumanMessage(
+      "What are the contents of the results.txt file?"
+    );
+
+    const secondResult = await model.invoke(
+      [firstMessage, firstResult, secondMessage],
+      {
+        container: container?.id, // Pass container to reuse files
+      }
+    );
+
+    // Verify second result succeeded
+    expect(secondResult).toBeInstanceOf(AIMessage);
+    const secondContent = Array.isArray(secondResult.content)
+      ? secondResult.content
+      : [];
+    expect(secondContent.length).toBeGreaterThan(0);
+
+    // Verify the same container was reused
+    const secondContainer = (secondResult as AIMessage).additional_kwargs
+      ?.container as { id: string; expires_at: string } | undefined;
+    expect(secondContainer?.id).toBe(container?.id);
+
+    expect(secondResult.text).toMatch(/30/);
+  }, 60000);
 });

libs/providers/langchain-anthropic/src/utils/extractGeneratedFiles.ts

@@ -0,0 +1,31 @@
+import type Anthropic from "@anthropic-ai/sdk";
+
+/**
+ * Extract generated file IDs from an Anthropic code execution response.
+ *
+ * Parses the response content blocks to find files created during code execution.
+ * Note: This function only returns file IDs.
+ *
+ * @param message - The Anthropic message response
+ * @returns Array of file IDs
+ */
+export function extractGeneratedFiles(
+  message: Anthropic.Beta.BetaMessage
+): string[] {
+  const fileIds: string[] = [];
+
+  for (const item of message.content) {
+    if (
+      item.type === "bash_code_execution_tool_result" &&
+      item.content.type === "bash_code_execution_result"
+    ) {
+      for (const file of item.content.content) {
+        if (file.type === "bash_code_execution_output") {
+          fileIds.push(file.file_id);
+        }
+      }
+    }
+  }
+
+  return fileIds;
+}

libs/providers/langchain-anthropic/src/utils/message_inputs.ts

@@ -81,6 +81,36 @@ function _formatImage(imageUrl: string) {
   );
 }
 
+/**
+ * Work around Anthropic API bug where bash_code_execution_output blocks
+ * cause 500 errors when included in conversation history.
+ * Filters out bash_code_execution_output blocks from bash_code_execution_tool_result blocks.
+ *
+ * @param contentPart The content block to potentially filter
+ */
+function _filterBashCodeExecutionOutputs(contentPart: { type: string }): void {
+  if (contentPart.type !== "bash_code_execution_tool_result") {
+    return;
+  }
+
+  const toolResult =
+    contentPart as unknown as Anthropic.Beta.Messages.BetaBashCodeExecutionToolResultBlockParam;
+
+  if (
+    toolResult.content.type === "bash_code_execution_result" &&
+    Array.isArray(toolResult.content.content)
+  ) {
+    // Filter out bash_code_execution_output blocks
+    const filteredContent = toolResult.content.content.filter(
+      (c) => c.type !== "bash_code_execution_output"
+    );
+    toolResult.content = {
+      ...toolResult.content,
+      content: filteredContent,
+    };
+  }
+}
+
 function _ensureMessageContents(messages: BaseMessage[]): BaseMessage[] {
   // Merge runs of human/tool messages into single human messages with content blocks.
   const updatedMsgs = [];
@@ -268,6 +298,9 @@ function* _formatContentBlocks(
           }
         }
       }
+
+      _filterBashCodeExecutionOutputs(contentPartCopy);
+
       // TODO: Fix when SDK types are fixed
       yield {
         ...contentPartCopy,