Add Shell Tool (#3335)

Create an official bash shell tool to replace the dynamically generated one
fix_agent_callbacks
Zander Chase 1 year ago committed by GitHub
parent 334c162f16
commit 5042bd40d3
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

@ -5,158 +5,158 @@
"id": "8f210ec3",
"metadata": {},
"source": [
"# Bash\n",
"It can often be useful to have an LLM generate bash commands, and then run them. A common use case for this is letting the LLM interact with your local file system. We provide an easy util to execute bash commands."
"# Shell Tool\n",
"\n",
"Giving agents access to the shell is powerful (though risky outside a sandboxed environment).\n",
"\n",
"The LLM can use it to execute any shell commands. A common use case for this is letting the LLM interact with your local file system."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "f7b3767b",
"metadata": {},
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"from langchain.utilities import BashProcess"
"from langchain.tools import ShellTool\n",
"\n",
"shell_tool = ShellTool()"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "cf1c92f0",
"metadata": {},
"outputs": [],
"source": [
"bash = BashProcess()"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "2fa952fc",
"metadata": {},
"id": "c92ac832-556b-4f66-baa4-b78f965dfba0",
"metadata": {
"tags": []
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"apify.ipynb\n",
"arxiv.ipynb\n",
"bash.ipynb\n",
"bing_search.ipynb\n",
"chatgpt_plugins.ipynb\n",
"ddg.ipynb\n",
"google_places.ipynb\n",
"google_search.ipynb\n",
"google_serper.ipynb\n",
"gradio_tools.ipynb\n",
"human_tools.ipynb\n",
"ifttt.ipynb\n",
"openweathermap.ipynb\n",
"python.ipynb\n",
"requests.ipynb\n",
"search_tools.ipynb\n",
"searx_search.ipynb\n",
"serpapi.ipynb\n",
"wikipedia.ipynb\n",
"wolfram_alpha.ipynb\n",
"zapier.ipynb\n",
"Hello World!\n",
"\n",
"real\t0m0.000s\n",
"user\t0m0.000s\n",
"sys\t0m0.000s\n",
"\n"
]
}
],
"source": [
"print(bash.run(\"ls\"))"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "e7896f8e",
"metadata": {},
"outputs": [
},
{
"name": "stdout",
"name": "stderr",
"output_type": "stream",
"text": [
"apify.ipynb\n",
"arxiv.ipynb\n",
"bash.ipynb\n",
"bing_search.ipynb\n",
"chatgpt_plugins.ipynb\n",
"ddg.ipynb\n",
"google_places.ipynb\n",
"google_search.ipynb\n",
"google_serper.ipynb\n",
"gradio_tools.ipynb\n",
"human_tools.ipynb\n",
"ifttt.ipynb\n",
"openweathermap.ipynb\n",
"python.ipynb\n",
"requests.ipynb\n",
"search_tools.ipynb\n",
"searx_search.ipynb\n",
"serpapi.ipynb\n",
"wikipedia.ipynb\n",
"wolfram_alpha.ipynb\n",
"zapier.ipynb\n",
"\n"
"/Users/wfh/code/lc/lckg/langchain/tools/shell/tool.py:34: UserWarning: The shell tool has no safeguards by default. Use at your own risk.\n",
" warnings.warn(\n"
]
}
],
"source": [
"bash.run(\"cd ..\")\n",
"# The commands are executed in a new subprocess each time, meaning that\n",
"# this call will return the same results as the last.\n",
"print(bash.run(\"ls\"))"
"print(shell_tool.run({\"commands\": [\"echo 'Hello World!'\", \"time\"]}))"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "851fee9f",
"id": "2fa952fc",
"metadata": {},
"source": [
"## Terminal Persistance\n",
"### Use with Agents\n",
"\n",
"By default, the bash command will be executed in a new subprocess each time. To retain a persistent bash session, we can use the `persistent=True` arg."
"As with all tools, these can be given to an agent to accomplish more complex tasks. Let's have the agent fetch some links from a web page."
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "4a93ea2c",
"metadata": {},
"outputs": [],
"source": [
"bash = BashProcess(persistent=True)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "a1e98b78",
"metadata": {},
"execution_count": 3,
"id": "851fee9f",
"metadata": {
"tags": []
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"custom_tools.ipynb\t\tmulti_input_tool.ipynb\n",
"examples\t\t\ttool_input_validation.ipynb\n",
"getting_started.md\n"
"\n",
"\n",
"\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\n",
"\u001b[32;1m\u001b[1;3mQuestion: What is the task?\n",
"Thought: We need to download the langchain.com webpage and extract all the URLs from it. Then we need to sort the URLs and return them.\n",
"Action:\n",
"```\n",
"{\n",
" \"action\": \"shell\",\n",
" \"action_input\": {\n",
" \"commands\": [\n",
" \"curl -s https://langchain.com | grep -o 'http[s]*://[^\\\" ]*' | sort\"\n",
" ]\n",
" }\n",
"}\n",
"```\n",
"\u001b[0m"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"/Users/wfh/code/lc/lckg/langchain/tools/shell/tool.py:34: UserWarning: The shell tool has no safeguards by default. Use at your own risk.\n",
" warnings.warn(\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"Observation: \u001b[36;1m\u001b[1;3mhttps://blog.langchain.dev/\n",
"https://discord.gg/6adMQxSpJS\n",
"https://docs.langchain.com/docs/\n",
"https://github.com/hwchase17/chat-langchain\n",
"https://github.com/hwchase17/langchain\n",
"https://github.com/hwchase17/langchainjs\n",
"https://github.com/sullivan-sean/chat-langchainjs\n",
"https://js.langchain.com/docs/\n",
"https://python.langchain.com/en/latest/\n",
"https://twitter.com/langchainai\n",
"\u001b[0m\n",
"Thought:\u001b[32;1m\u001b[1;3mThe URLs have been successfully extracted and sorted. We can return the list of URLs as the final answer.\n",
"Final Answer: [\"https://blog.langchain.dev/\", \"https://discord.gg/6adMQxSpJS\", \"https://docs.langchain.com/docs/\", \"https://github.com/hwchase17/chat-langchain\", \"https://github.com/hwchase17/langchain\", \"https://github.com/hwchase17/langchainjs\", \"https://github.com/sullivan-sean/chat-langchainjs\", \"https://js.langchain.com/docs/\", \"https://python.langchain.com/en/latest/\", \"https://twitter.com/langchainai\"]\u001b[0m\n",
"\n",
"\u001b[1m> Finished chain.\u001b[0m\n"
]
},
{
"data": {
"text/plain": [
"'[\"https://blog.langchain.dev/\", \"https://discord.gg/6adMQxSpJS\", \"https://docs.langchain.com/docs/\", \"https://github.com/hwchase17/chat-langchain\", \"https://github.com/hwchase17/langchain\", \"https://github.com/hwchase17/langchainjs\", \"https://github.com/sullivan-sean/chat-langchainjs\", \"https://js.langchain.com/docs/\", \"https://python.langchain.com/en/latest/\", \"https://twitter.com/langchainai\"]'"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"bash.run(\"cd ..\")\n",
"# Note the list of files is different\n",
"print(bash.run(\"ls\"))"
"from langchain.chat_models import ChatOpenAI\n",
"from langchain.agents import initialize_agent\n",
"from langchain.agents import AgentType\n",
"\n",
"llm = ChatOpenAI(temperature=0)\n",
"\n",
"shell_tool.description = shell_tool.description + f\"args {shell_tool.args}\".replace(\"{\", \"{{\").replace(\"}\", \"}}\")\n",
"self_ask_with_search = initialize_agent([shell_tool], llm, agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION, verbose=True)\n",
"self_ask_with_search.run(\"Download the langchain.com webpage and grep for all urls. Return only a sorted list of them. Be sure to use double quotes.\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e13c1c9c",
"id": "8d0ea3ac-0890-4e39-9cec-74bd80b4b8b8",
"metadata": {},
"outputs": [],
"source": []

@ -27,6 +27,7 @@ from langchain.tools.requests.tool import (
RequestsPutTool,
)
from langchain.tools.searx_search.tool import SearxSearchResults, SearxSearchRun
from langchain.tools.shell.tool import ShellTool
from langchain.tools.wikipedia.tool import WikipediaQueryRun
from langchain.tools.wolfram_alpha.tool import WolframAlphaQueryRun
from langchain.utilities import ArxivAPIWrapper
@ -67,11 +68,7 @@ def _get_tools_requests_delete() -> BaseTool:
def _get_terminal() -> BaseTool:
return Tool(
name="Terminal",
description="Executes commands in a terminal. Input should be valid commands, and the output will be any output from running that command.",
func=BashProcess().run,
)
return ShellTool()
_BASE_TOOLS: Dict[str, Callable[[], BaseTool]] = {

@ -26,9 +26,11 @@ from langchain.tools.playwright import (
NavigateTool,
)
from langchain.tools.plugin import AIPluginTool
from langchain.tools.shell.tool import ShellTool
__all__ = [
"AIPluginTool",
"APIOperation",
"BaseBrowserTool",
"BaseTool",
"BaseTool",
@ -55,6 +57,6 @@ __all__ = [
"NavigateTool",
"OpenAPISpec",
"ReadFileTool",
"ShellTool",
"WriteFileTool",
"APIOperation",
]

@ -0,0 +1,5 @@
"""Shell tool."""
from langchain.tools.shell.tool import ShellTool
__all__ = ["ShellTool"]

@ -0,0 +1,71 @@
import asyncio
import platform
import warnings
from typing import List, Type
from pydantic import BaseModel, Field, root_validator
from langchain.tools.base import BaseTool
from langchain.utilities.bash import BashProcess
class ShellInput(BaseModel):
"""Commands for the Bash Shell tool."""
commands: List[str] = Field(
...,
description="List of shell commands to run. Deserialized using json.loads",
)
"""List of shell commands to run."""
@root_validator
def _validate_commands(cls, values: dict) -> dict:
"""Validate commands."""
# TODO: Add real validators
commands = values.get("commands")
if not isinstance(commands, list):
values["commands"] = [commands]
# Warn that the bash tool is not safe
warnings.warn(
"The shell tool has no safeguards by default. Use at your own risk."
)
return values
def _get_default_bash_processs() -> BashProcess:
"""Get file path from string."""
return BashProcess(return_err_output=True)
def _get_platform() -> str:
"""Get platform."""
system = platform.system()
if system == "Darwin":
return "MacOS"
return system
class ShellTool(BaseTool):
"""Tool to run shell commands."""
process: BashProcess = Field(default_factory=_get_default_bash_processs)
"""Bash process to run commands."""
name: str = "terminal"
"""Name of tool."""
description: str = f"Run shell commands on this {_get_platform()} machine."
"""Description of tool."""
args_schema: Type[BaseModel] = ShellInput
"""Schema for input arguments."""
def _run(self, commands: List[str]) -> str:
"""Run commands and return final output."""
return self.process.run(commands)
async def _arun(self, commands: List[str]) -> str:
"""Run commands asynchronously and return final output."""
return await asyncio.get_event_loop().run_in_executor(
None, self.process.run, commands
)

@ -0,0 +1,43 @@
import warnings
import pytest
from langchain.tools.shell.tool import ShellInput, ShellTool
# Test data
test_commands = ["echo 'Hello, World!'", "echo 'Another command'"]
def test_shell_input_validation() -> None:
shell_input = ShellInput(commands=test_commands)
assert isinstance(shell_input.commands, list)
assert len(shell_input.commands) == 2
with warnings.catch_warnings(record=True) as w:
ShellInput(commands=test_commands)
assert len(w) == 1
assert (
str(w[-1].message)
== "The shell tool has no safeguards by default. Use at your own risk."
)
def test_shell_tool_init() -> None:
shell_tool = ShellTool()
assert shell_tool.name == "terminal"
assert isinstance(shell_tool.description, str)
assert shell_tool.args_schema == ShellInput
assert shell_tool.process is not None
@pytest.mark.asyncio
async def test_shell_tool_arun() -> None:
shell_tool = ShellTool()
result = await shell_tool._arun(commands=test_commands)
assert result.strip() == "Hello, World!\nAnother command"
def test_shell_tool_run() -> None:
shell_tool = ShellTool()
result = shell_tool._run(commands=test_commands)
assert result.strip() == "Hello, World!\nAnother command"
Loading…
Cancel
Save