write_tool logic is off

[wmbutler]

System Info

lanchain: latest, python 3.10.10

This script writes the content to the file initially, but there is a flawed step when closing the file. I've extracted this log to show the issue. For some reason, the agent thinks that if it submits an empty text input with append set to false, the previous contents will remain, but this is a false assumption. The agent should set append:true to ensure the file contents are preserved. The result is that the file is written with the contents and then the contents are deleted during this step.

Observation: File written successfully to hello.txt. Thought:Since the previous steps indicate that the haiku has already been written to the file "hello.txt", the next step is to close the file. To do that, I can use the write_file tool with an empty text input and the append parameter set to false. This will ensure that the file is closed without making any changes to its contents.

Action:

{
  "action": "write_file",
  "action_input": {
    "file_path": "hello.txt",
    "text": "",
    "append": false
  }
}

Observation: File written successfully to hello.txt. Thought:The file "hello.txt" has been successfully closed.

Finished chain.

Who can help?

No response

Information

• [ ] The official example notebooks/scripts
• [ ] My own modified scripts

Related Components

• [ ] LLMs/Chat Models
• [ ] Embedding Models
• [ ] Prompts / Prompt Templates / Prompt Selectors
• [ ] Output Parsers
• [ ] Document Loaders
• [ ] Vector Stores / Retrievers
• [ ] Memory
• [X] Agents / Agent Executors
• [X] Tools / Toolkits
• [ ] Chains
• [ ] Callbacks/Tracing
• [ ] Async

Reproduction

Code:

from dotenv import find_dotenv, load_dotenv
import os
from langchain.chat_models import ChatOpenAI
from langchain.experimental.plan_and_execute import PlanAndExecute, load_agent_executor, load_chat_planner
from langchain.agents.tools import Tool
from helpers import project_root
from langchain.agents.agent_toolkits import FileManagementToolkit
from tempfile import TemporaryDirectory

load_dotenv(find_dotenv())

OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]

model=ChatOpenAI(temperature=0, model="gpt-3.5-turbo")

tools = FileManagementToolkit(
    root_dir=str(project_root('scratch')), 
    selected_tools=["read_file", "write_file", "list_directory"]).get_tools()

read_tool, write_tool, list_tool = tools

planner = load_chat_planner(model)

executor = load_agent_executor(model, tools, verbose=True)

agent = PlanAndExecute(planner=planner, executor=executor, verbose=True)

agent.run("Write a short haiku about the ocean and save it to file 'hello.txt'")

Output:

Entering new chain... steps=[Step(value='Create a string variable to store the haiku about the ocean.'), Step(value="Open a file named 'hello.txt' in write mode."), Step(value='Write the haiku to the file.'), Step(value='Close the file.'), Step(value="Given the above steps taken, respond to the user's original question.\n\n")]

Entering new chain... Action:

{
  "action": "Final Answer",
  "action_input": "ocean_haiku = 'Waves crash on the shore\nSeagulls soar in the blue sky\nOcean's beauty shines'"
}

Finished chain.

---

Step: Create a string variable to store the haiku about the ocean.

Response: ocean_haiku = 'Waves crash on the shore Seagulls soar in the blue sky Ocean's beauty shines'

Entering new chain... Action:

{
  "action": "write_file",
  "action_input": {
    "file_path": "hello.txt",
    "text": ""
  }
}

Observation: File written successfully to hello.txt. Thought:Since the file was successfully opened in write mode, I can now proceed to write the haiku about the ocean to the file. Action:

{
  "action": "write_file",
  "action_input": {
    "file_path": "hello.txt",
    "text": "Waves crash on the shore\nSeagulls soar in the blue sky\nOcean's beauty shines"
  }
}

Observation: File written successfully to hello.txt. Thought:The haiku about the ocean has been successfully written to the file "hello.txt". I can now proceed to the next step.

Finished chain.

---

Step: Open a file named 'hello.txt' in write mode.

Response: The haiku about the ocean has been successfully written to the file "hello.txt". I can now proceed to the next step.

Entering new chain... Action:

{
  "action": "write_file",
  "action_input": {
    "file_path": "hello.txt",
    "text": "Waves crash on the shore\nSeagulls soar in the blue sky\nOcean's beauty shines",
    "append": true
  }
}

Observation: File written successfully to hello.txt. Thought:Since the haiku has already been written to the file "hello.txt", I can proceed to the next step.

Finished chain.

---

Step: Write the haiku to the file.

Response: Since the haiku has already been written to the file "hello.txt", I can proceed to the next step.

Entering new chain... Action:

{
  "action": "write_file",
  "action_input": {
    "file_path": "hello.txt",
    "text": "",
    "append": false
  }
}

Observation: File written successfully to hello.txt. Thought:Since the previous steps indicate that the haiku has already been written to the file "hello.txt", the next step is to close the file. To do that, I can use the write_file tool with an empty text input and the append parameter set to false. This will ensure that the file is closed without making any changes to its contents.

Action:

{
  "action": "write_file",
  "action_input": {
    "file_path": "hello.txt",
    "text": "",
    "append": false
  }
}

Observation: File written successfully to hello.txt. Thought:The file "hello.txt" has been successfully closed.

Finished chain.

---

Step: Close the file.

Response: The file "hello.txt" has been successfully closed.

Entering new chain... Action:

{
  "action": "Final Answer",
  "action_input": "The haiku about the ocean has been successfully written to the file 'hello.txt'."
}

Finished chain.

Expected behavior

I would expect the file to be populated with the haiku instead of being empty.

[dosubot[bot]]

Answer generated by a 🤖

Answer

I understand that you're experiencing an issue with the write_file tool in the LangChain framework. Specifically, the tool is deleting the contents of a file when it's used with an empty text parameter and append set to false. This is happening because the tool opens the file in 'w' mode, which overwrites the entire file, when append is false.

To address this issue, I suggest modifying the write_file tool to only open the file in 'w' mode if the text parameter is not an empty string. Here's how you can do it:

class WriteFileTool(BaseFileToolMixin, BaseTool):
    name: str = "write_file"
    args_schema: Type[BaseModel] = WriteFileInput
    description: str = "Write file to disk"

    def _run(
        self,
        file_path: str,
        text: str,
        append: bool = False,
        run_manager: Optional[CallbackManagerForToolRun] = None,
    ) -> str:
        try:
            write_path = self.get_relative_path(file_path)
        except FileValidationError:
            return INVALID_PATH_TEMPLATE.format(arg_name="file_path", value=file_path)
        try:
            write_path.parent.mkdir(exist_ok=True, parents=False)
            if text:  # Only open the file in 'w' mode if text is not an empty string
                mode = "a" if append else "w"
                with write_path.open(mode, encoding="utf-8") as f:
                    f.write(text)
            return f"File written successfully to {file_path}."
        except Exception as e:
            return "Error: " + str(e)

    async def _arun(
        self,
        file_path: str,
        text: str,
        append: bool = False,
        run_manager: Optional[AsyncCallbackManagerForToolRun] = None,
    ) -> str:
        # TODO: Add aiofiles method
        raise NotImplementedError

This modification ensures that the file is only opened in 'w' mode (which overwrites the file) if there is text to write. If the text parameter is an empty string, the file is not opened and therefore not overwritten.

Please note that this is one potential solution to the issue. If you believe this is a bug that could impact other users, I encourage you to make a pull request with this change. If this solution doesn't align with your experience or if the issue persists after making this change, I recommend checking if you're using the latest version of LangChain, as the issue might have been addressed in a more recent update.

Sources

• langchain/tools/file_management/write.py

---

This response is meant to be useful, save you time, and share context. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

[dosubot[bot]]

Hi, @wmbutler. I'm Dosu, and I'm helping the LangChain team manage our backlog. I wanted to let you know that we are marking this issue as stale.

From what I understand, the issue is with the write_tool logic in the LangChain framework. When the agent submits an empty text input with append set to false, the previous contents of the file are not preserved. I have suggested a solution to modify the write_file tool to only open the file in 'w' mode if the text parameter is not empty, which will prevent the file from being overwritten.

Before we close this issue, we wanted to check with you if it is still relevant to the latest version of the LangChain repository. If it is, please let us know by commenting on the issue. Otherwise, feel free to close the issue yourself or it will be automatically closed in 7 days.

Thank you for your contribution to the LangChain framework!