Skip to content

armgit / camel

Public Imported from GitHub
0

🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org

9.6 KB
Raw Blame
# ========= Copyright 2023-2026 @ CAMEL-AI.org. All Rights Reserved. =========
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ========= Copyright 2023-2026 @ CAMEL-AI.org. All Rights Reserved. =========
import json
from typing import Dict, List

import pytest

from camel.memories import ContextRecord
from camel.messages import (
    BaseMessage,
    FunctionCallingMessage,
    HermesFunctionFormatter,
)
from camel.models import ModelFactory
from camel.societies import RolePlaying
from camel.toolkits import MathToolkit
from camel.types import ModelPlatformType, ModelType, RoleType, TaskType


@pytest.fixture
def assistant_func_call_message() -> FunctionCallingMessage:
    role_name = "assistant"
    role_type = RoleType.ASSISTANT
    meta_dict = None
    content = "test function message"

    return FunctionCallingMessage(
        role_name=role_name,
        role_type=role_type,
        meta_dict=meta_dict,
        content=content,
        func_name="add",
        args={"a": "1", "b": "2"},
        tool_call_id=None,
    )


@pytest.fixture
def function_result_message() -> FunctionCallingMessage:
    role_name = "function"
    role_type = RoleType.ASSISTANT
    meta_dict = None

    return FunctionCallingMessage(
        role_name=role_name,
        role_type=role_type,
        meta_dict=meta_dict,
        content="",
        func_name="add",
        result=3,
        tool_call_id=None,
    )


def test_assistant_func_message(
    assistant_func_call_message: FunctionCallingMessage,
):
    content = "test function message"

    assert assistant_func_call_message.func_name == "add"
    assert assistant_func_call_message.args == {"a": "1", "b": "2"}

    result = assistant_func_call_message.to_openai_assistant_message()
    assert result["role"] == "assistant"
    assert result["content"] == content
    assert len(result["tool_calls"]) == 1  # type: ignore[arg-type]
    tool_call = result["tool_calls"][0]  # type: ignore[index]
    assert tool_call["type"] == "function"
    assert tool_call["function"]["name"] == "add"
    assert tool_call["function"]["arguments"] == json.dumps(
        {"a": "1", "b": "2"}
    )


def test_function_func_message(
    function_result_message: FunctionCallingMessage,
):
    assert function_result_message.func_name == "add"
    assert function_result_message.result == 3

    msg_dict: Dict[str, str] = {
        "role": "tool",
        "content": str(3),
        "tool_call_id": "null",
    }
    assert function_result_message.to_openai_tool_message() == msg_dict


def test_assistant_func_message_to_openai_tool_message(
    assistant_func_call_message: FunctionCallingMessage,
):
    expected_msg_dict: Dict[str, str] = {
        "role": "tool",
        "content": str(None),
        "tool_call_id": "null",
    }

    assert (
        assistant_func_call_message.to_openai_tool_message()
        == expected_msg_dict
    )


@pytest.mark.model_backend
def test_roleplay_conversion_with_tools():
    tools = MathToolkit().get_tools()
    model = ModelFactory.create(
        model_platform=ModelPlatformType.OPENAI,
        model_type=ModelType.GPT_4O_MINI,
    )

    role_playing = RolePlaying(
        assistant_role_name="assistant",
        assistant_agent_kwargs=dict(
            model=model,
            tools=tools,
        ),
        user_role_name="user",
        user_agent_kwargs=dict(model=model),
        task_prompt="Perform the task",
        task_specify_agent_kwargs=dict(model=model),
        task_type=TaskType.AI_SOCIETY,
    )
    input_msg = role_playing.init_chat("What is 2 + 4?")
    [assistant, _] = role_playing.step(input_msg)
    role_playing.step(assistant.msg)

    records: List[ContextRecord] = (
        role_playing.assistant_agent.memory.retrieve()
    )
    original_messages = []
    sharegpt_msgs = []

    for record in records:
        message = record.memory_record.message
        # Remove meta_dict to avoid comparison issues
        message.meta_dict = None
        # Clear tool_call_id for function messages
        if isinstance(message, FunctionCallingMessage):
            message.tool_call_id = ""
        original_messages.append(message)
        sharegpt_msgs.append(message.to_sharegpt())

    converted_back = []
    for msg in sharegpt_msgs:
        message = BaseMessage.from_sharegpt(
            msg, function_format=HermesFunctionFormatter()
        )
        # Clear tool_call_id for function messages
        if isinstance(message, FunctionCallingMessage):
            message.tool_call_id = ""
        converted_back.append(message)

    assert converted_back == original_messages


def test_convert_function_call_and_response_to_from_sharegpt_hermes(
    assistant_func_call_message: FunctionCallingMessage,
    function_result_message: FunctionCallingMessage,
):
    sharegpt_function_call = assistant_func_call_message.to_sharegpt()

    # Check the function call contains a hermes function call
    # TODO: Consider using code from https://github.com/NousResearch/Hermes-Function-Calling/blob/main/validator.py # noqa: E501
    #  and adjacent files
    assert "<tool_call>" in sharegpt_function_call.value

    # Test it converts back
    reconverted_function_call = BaseMessage.from_sharegpt(
        sharegpt_function_call
    )
    assert assistant_func_call_message == reconverted_function_call

    sharegpt_function_result = function_result_message.to_sharegpt()
    reconverted_function_result = BaseMessage.from_sharegpt(
        sharegpt_function_result
    )

    # Set reference function call to take on CAMEL function result role
    function_result_message.role_name = "assistant"
    assert function_result_message == reconverted_function_result


def test_hermes_sharegpt_roundtrip_preserves_non_string_args():
    r"""Tool-call args with apostrophes, booleans, None and floats must
    survive a ``to_sharegpt()`` -> ``from_sharegpt()`` round-trip.

    They previously did not: the formatter serialized args with Python
    ``repr`` and parsed them by swapping ``'`` for ``"``, which corrupts any
    string containing an apostrophe and rejects ``True``/``False``/``None``.
    The tool call then failed to parse and silently degraded to a plain
    ``BaseMessage`` with the arguments lost.
    """
    args = {
        "note": "it's sunny",
        "active": True,
        "missing": None,
        "ratio": 0.5,
        "city": "London",
    }
    message = FunctionCallingMessage(
        role_name="assistant",
        role_type=RoleType.ASSISTANT,
        meta_dict=None,
        content="",
        func_name="note_tool",
        args=args,
        tool_call_id=None,
    )

    sharegpt = message.to_sharegpt()
    reconverted = BaseMessage.from_sharegpt(
        sharegpt, function_format=HermesFunctionFormatter()
    )

    assert isinstance(reconverted, FunctionCallingMessage)
    assert reconverted.func_name == "note_tool"
    assert reconverted.args == args


def test_hermes_extract_tool_calls_parses_legacy_repr_payload():
    r"""Legacy single-quoted Python-``repr`` ``<tool_call>`` payloads (as
    emitted by earlier CAMEL versions) must still parse, so previously
    serialized data keeps round-tripping."""
    formatter = HermesFunctionFormatter()
    legacy = (
        "<tool_call>\n"
        "{'name': 'note_tool', 'arguments': {'city': 'London', "
        "'active': True}}\n"
        "</tool_call>"
    )

    calls = formatter.extract_tool_calls(legacy)

    assert len(calls) == 1
    assert calls[0].name == "note_tool"
    assert calls[0].arguments == {"city": "London", "active": True}


def test_function_func_message_to_openai_assistant_message(
    function_result_message: FunctionCallingMessage,
):
    with pytest.raises(
        ValueError,
        match=(
            "Invalid request for converting into assistant message"
            " due to missing function name or arguments."
        ),
    ):
        function_result_message.to_openai_assistant_message()


def test_masking_in_openai_tool_message():
    msg = FunctionCallingMessage(
        role_name="assistant",
        role_type=RoleType.ASSISTANT,
        meta_dict=None,
        content="",
        func_name="get_user_data",
        result={"user_id": "123", "secret": "abc"},
        tool_call_id="tool123",
        mask_output=True,
    )

    openai_msg = msg.to_openai_tool_message()
    assert openai_msg["role"] == "tool"
    assert openai_msg["tool_call_id"] == "tool123"
    assert openai_msg["content"] == "[MASKED]"


def test_masking_in_sharegpt():
    msg = FunctionCallingMessage(
        role_name="assistant",
        role_type=RoleType.ASSISTANT,
        meta_dict=None,
        content="",
        func_name="get_user_data",
        result={"user_id": "123", "secret": "abc"},
        tool_call_id="tool456",
        mask_output=True,
    )

    sharegpt_msg = msg.to_sharegpt()
    assert sharegpt_msg.value == "[MASKED]"


def test_to_dict_includes_mask_output():
    msg = FunctionCallingMessage(
        role_name="assistant",
        role_type=RoleType.ASSISTANT,
        meta_dict=None,
        content="Hello",
        func_name="do_stuff",
        result={"status": "ok"},
        tool_call_id="tool789",
        mask_output=True,
    )

    d = msg.to_dict()
    assert d["mask_output"] is True
    assert d["content"] == "Hello"