@langchain/aws Bug: ChatBedrockConverse fails when calling multiple tools simultaneously

[tinque]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain.js documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain.js rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

I use the legacy createToolCallingAgent and AgentExecutor from "langchain/agents", but I encounter the same problem with the LangGraph agent.

import { ChatBedrockConverse } from "@langchain/aws";
import { tool } from "@langchain/core/tools";
import { ChatPromptTemplate } from "@langchain/core/prompts";
import { createToolCallingAgent, AgentExecutor } from "langchain/agents";
import { z } from "zod";

const model = new ChatBedrockConverse({
  model: "anthropic.claude-3-sonnet-20240229-v1:0",
  region: "us-east-1",
});

const weatherTool = tool(
  ({ city, state }) => `The weather in ${city}, ${state} is 72°F and sunny`,
  {
    name: "weather_tool",
    description: "Get the weather for a city",
    schema: z.object({
      city: z.string().describe("The city to get the weather for"),
      state: z.string().describe("The state to get the weather for").optional(),
    }),
  }
);

const tools = [weatherTool];

const modelWithTools = model.bindTools(tools);


const prompt = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant"],
  ["placeholder", "{chat_history}"],
  ["human", "{input}"],
  ["placeholder", "{agent_scratchpad}"],
]);

const agent = await createToolCallingAgent({ llm: model, tools, prompt });

const agentExecutor = new AgentExecutor({
  agent,
  tools,
});

const res = await agentExecutor.invoke({
  input: "What's the weather in New York and Paris?",
})
console.log(res);

Error Message and Stack Trace (if applicable)

/Users/quentin/contrib/repro_bug/node_modules/@aws-sdk/client-bedrock-runtime/dist-cjs/index.js:1159
  const exception = new ValidationException({
                    ^

ValidationException: The number of toolResult blocks at messages.2.content exceeds the number of toolUse blocks of previous turn.
    at de_ValidationExceptionRes (/Users/quentin/contrib/repro_bug/node_modules/@aws-sdk/client-bedrock-runtime/dist-cjs/index.js:1159:21)
    at de_CommandError (/Users/quentin/contrib/repro_bug/node_modules/@aws-sdk/client-bedrock-runtime/dist-cjs/index.js:1008:19)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async /Users/quentin/contrib/repro_bug/node_modules/@smithy/middleware-serde/dist-cjs/index.js:35:20
    at async /Users/quentin/contrib/repro_bug/node_modules/@smithy/core/dist-cjs/index.js:165:18
    at async /Users/quentin/contrib/repro_bug/node_modules/@smithy/middleware-retry/dist-cjs/index.js:320:38
    at async /Users/quentin/contrib/repro_bug/node_modules/@aws-sdk/middleware-logger/dist-cjs/index.js:34:22
    at async ChatBedrockConverse._streamResponseChunks (file:///Users/quentin/contrib/repro_bug/node_modules/@langchain/aws/dist/chat_models.js:247:26)
    at async ChatBedrockConverse._streamIterator (file:///Users/quentin/contrib/repro_bug/node_modules/@langchain/core/dist/language_models/chat_models.js:89:34)
    at async ChatBedrockConverse.transform (file:///Users/quentin/contrib/repro_bug/node_modules/@langchain/core/dist/runnables/base.js:387:9) {
  '$fault': 'client',
  '$metadata': {
    httpStatusCode: 400,
    requestId: 'd92de5a1-fa53-42dd-88fd-d0438fafe27c',
    extendedRequestId: undefined,
    cfId: undefined,
    attempts: 1,
    totalRetryDelay: 0
  }
}

Node.js v20.12.0

### Description

When the LLM calls more than one tool simultaneously, the following error is encountered:

ValidationException: The number of toolResult blocks at messages.2.content exceeds the number of toolUse blocks of previous turn.

The issue seems to originate from the fact that each `ToolMessage` is converted into a `user` type message with a single `toolResult` block (see the relevant code [here](https://github.com/langchain-ai/langchainjs/blob/main/libs/langchain-aws/src/common.ts#L170)). However, Bedrock expects a single message containing multiple `toolResult` blocks in the content.

For more details, refer to the AWS documentation:
- [API_runtime_Message](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Message.html)
- [API_runtime_ContentBlock](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ContentBlock.html)


### System Info

langchain@0.2.10
platform mac
node v20.12.0
npm 10.5.0

[bracesproul]

Thank you for flagging this! I'll take a look tomorrow and update you with my findings.

[tinque]

I've proposed a fix, but I find it a bit messy... I don't have any other ideas.