Agents 2.3 -> 3.0

With Mythic 3.0, you have two options for your containers - GoLang or Python. This section will only go over the Python side since there are no 2.3 agents with GoLang containers.

When you git clone the new Mythic v3.0.0 you'll notice that there's no mythic-cli binary. To reduce the size of the GitHub clones, this binary is now included as part of the main base docker image, so run sudo make and the binary will be downloaded and copied into the normal spot.

The Apfell v3.0.0 (https://github.com/MythicAgents/apfell/tree/v3.0.0) branch and the Apollo v3.0.0 (https://github.com/MythicAgents/apollo/tree/v3.0.0) branch are good examples of two slightly different ways you can format things.

Make sure you update your Docker version to at least 20.10.22 or above (current is 23.0.1. This is required for the latest docker containers to work properly. A simple sudo apt upgrade and install should suffice. Also install docker-compose via sudo apt install docker-compose-plugin vs the docker-compose script as the script will soon be deprecated according to Docker.

The ExternalAgent format is still the same. These next pieces are how you can test updates of your agent locally before copying your agent's folder back into your normal ExternalAgent format.

The Mythic/InstalledServices folder is on the Mythic server. When you installed a PayloadType via ./mythic-cli install github <url>, the folder within your github's Payload_Type folder is copied to the Mythic/InstalledServices folder.

If you are doing local agent development instead of on the same server as the Mythic instance, then in the following steps you can treat the Mythic/InstalledServices folder the same as your GitHub's Payload_Type folder (or any other folder really - it's simply serving as a staging ground while you create the new folder structure).

If you're doing local development, you need at least Python version 3.10 because of the new typing features it offers.

  1. Make a directory, agent name , in Mythic/InstalledServices

  2. Copy your entire Payload type's agent name directory into InstalledServices (yes, the path will look like Mythic/InstalledServices/agentName/agentName)

  3. In Mythic/InstalledServices/agentName create a main.py and a Dockerfile

  4. In your new Dockerfile, copy the contents of your old Dockerfile and change the FROM line to FROM itsafeaturemythic/mythic_python_base:latest

  5. In your new main.py add:

    import mythic_container
    from [agent name].mythic import *
    mythic_container.mythic_service.start_and_run_forever()
  6. To make your new agent name directory a PyPi package that can be imported, create a __init__.py file in Mythic/InstalledServices/agentName/agentName. In the Mythic/InstalledServices/agentName/agentName/Mythic folder make a __init__.py file with the following contents (this will loop through all of your command files and import them automatically):

import glob
import os.path
from pathlib import Path
from importlib import import_module, invalidate_caches
import sys
# Get file paths of all modules.

currentPath = Path(__file__)
searchPath = currentPath.parent / "agent_functions" / "*.py"
modules = glob.glob(f"{searchPath}")
invalidate_caches()
for x in modules:
    if not x.endswith("__init__.py") and x[-3:] == ".py":
        module = import_module(f"{__name__}.agent_functions." + Path(x).stem)
        for el in dir(module):
            if "__" not in el:
                globals()[el] = getattr(module, el)


sys.path.append(os.path.abspath(currentPath.name))
  1. In your Mythic/InstalledServices/agentName/agentName/mythic/agent_functions files, we need to replace all mythic_payloadtype_container with mythic_container . If you have an import like from agent_functions.execute_pe import PRINTSPOOFER_FILE_ID which references another command file, update it to from .execute_pe import PRINTSPOOFER_FILE_ID. If you include a local library at the same level as agent_functions, you can import it like from [agent name].mythic.[package] import [thing]

If you're doing local development, you'll need a rabbitmq_config.json file at the same level as your main.py to tell your service where Mythic is located and the rabbitmq password. The configuration options you can supply can be found in the Local Development section.

There are some changes to the rabbitmq_config.json file keys:

  • container_files_path is no longer used and can be deleted.

  • username is no longer used and can be deleted.

  • password is now rabbitmq_password.

  • host is now rabbitmq_host.

  • name is no longer used and can be deleted.

  • virtual_host is no longer used and can be deleted.

Required keys in rabbitmq_config.json are:

  • rabbitmq_host - points to the IP where Mythic lives

  • rabbitmq_password - the password used to authenticate to rabbitmq

  • mythic_server_host - points to the IP where Mythic lives

  • mythic_server_port - if you're using something other than the default (this is NOT the 7443 that you use for the UI)

  • mythic_server_grpc_port - if you're using something other than the default

Now to actually update the content of your builder/command files. There's not much you need to do.

builder.py agent definition

Because the new structure treats your entire agent directory as a Python package, the container no longer knows the paths for things. This gives you a lot more freedom in how you want to organize your code, but does require you to specify where things are located. In your builder.py file where you define your Payload Type, you need to add the following:

agent_path = pathlib.Path(".") / "apollo" / "mythic"
agent_code_path = pathlib.Path(".") / "apollo" / "agent_code"
agent_icon_path = agent_path / "agent_functions" / "apollo.svg"

The agent_path is the path to your general agent structure (typically with the agent_functions as a sub-folder. The agent_code_path points to your agent's actual code.

Something that's a little different is the agent icons - the agents will sync that over automatically with the rest of their definition (no more having to copy it over manually or get it from an install). What that means though is you either need to supply agent_icon_path and provide the path to your agent's svg icon or specify agent_icon_bytes and provide the raw bytes for your icon.

build

In your payload type's build function you can report back on build steps via the SendMythicRPCPayloadUpdateBuildStep RPC call (based on your defined build steps). This will update the UI step-by-step for the operator so they know what's going on.

You can also set UpdatedFilename (or updated_filename for Python) in your build response and adjust the final filename of the payload. This can be helpful if your payload type allows you to build to various outputs (exe, dll, dylib, binary, etc). This allows you to adjust the filename based on that so that when the user clicks "download" in the UI, they get the right file and don't have to change the filename.

Browser scripts

Browserscripts work just the same, but browserscripts will look for their code at agent_path / browser_scripts / filename.js OR at the path specified by the name parameter for the script. So, that means your can either specify the name as test.js and have it located in your agent_path / browser_scripts / test.js file or specify a full path as your name.

The browser_script attribute is a single BrowserScript value, not an array. This is because the entire Python back-end is gone, so there's no more need to supply a script for the old UI and the new UI.

c2 profile parameters when building

When looping through c2 profile parameters - arrays are actually arrays, crypto types and dictionary types are dictionaries, so do better checking here for name of parameters. A bunch of agents simply check if the supplied value is a dictionary and then automatically try to pull out certain values, but that might not be the case anymore. For example, when looping through the http profile, both the AESPSK and the headers parameters will be passed in as dictionaries.

for key, val in c2.get_parameters_dict().items():
    if key == "AESPSK":
        c2_code = c2_code.replace(key, val["enc_key"] if val["enc_key"] is not None else "")
    elif not isinstance(val, str):
        c2_code = c2_code.replace(key, json.dumps(val))
    else:
        c2_code = c2_code.replace(key, val)

Updated Create Tasking - create_go_tasking

The current create_tasking functions still work just like normal; however, the newer create_go_tasking function gives you more contextual data and mirrors the data structures from the new Golang container version.

async def create_go_tasking(self, taskData: MythicCommandBase.PTTaskMessageAllData) -> MythicCommandBase.PTTaskCreateTaskingMessageResponse:
    response = MythicCommandBase.PTTaskCreateTaskingMessageResponse(
        TaskID=taskData.Task.ID,
        Success=True,
    )
    return response

This taskData variable is defined here: https://github.com/MythicMeta/MythicContainerPyPi/blob/main/mythic_container/MythicCommandBase.py#L1068 and provides a lot more context in a well-defined class.

Completion Function for Tasking

Tasks can specify for a certain function to execute when the task finishes executing. That hasn't changed. However, the format of how you define it has changed slightly. Before, you'd simply pass the name of a function and the container would loop through all known function definitions looking for one that matched. That's not super great, so now you define a dictionary of function name to function as part of your command definition.

completion_functions: dict[str, Callable[[PTTaskCompletionFunctionMessage], Awaitable[PTTaskCompletionFunctionMessageResponse]]] = {}

The PTTaskCompletionFunctionMessage and response classes can be found in the PyPi code and auto-completed via IDEs. This syntax is just the Python way of saying that the format is:\

async def functionName(myArg: PTTaskCompletionFunctionMessage) -> PTTaskCompletionFunctionMessageResponse:
    do something here

To leverage this new functionName function as part of your tasking, in your create_tasking function you need to set the name:

async def create_tasking(self, task: MythicTask) -> MythicTask:
        task.completed_callback_function = "functionName"
        return task

If you're using the new create_go_tasking function, then you need to do somthing very similar:

async def create_go_tasking(self, taskData: MythicCommandBase.PTTaskMessageAllData) -> MythicCommandBase.PTTaskCreateTaskingMessageResponse:
    response = MythicCommandBase.PTTaskCreateTaskingMessageResponse(
        TaskID=taskData.Task.ID,
        CompletionFunctionName="functionName"
    )
    return response

Process response function

Sending back data via the process_response key within your responses allows you to hook into the associated command's process_response function within your Payload Type's container. The format of this function has changed slightly:

old:

async def process_response(self, response: AgentResponse):
    resp = await MythicRPC().execute("update_callback", task_id=response.task.id, sleep_info=response.response)

new:

async def process_response(self, task: PTTaskMessageAllData, response: any) -> PTTaskProcessResponseMessageResponse:
        resp = PTTaskProcessResponseMessageResponse(TaskID=task.Task.ID, Success=True)
        await MythicRPC().execute("update_callback", task_id=task.Task.ID, sleep_info=response)
        return resp

Dynamic Query Function

Similar to the completion functions, dynamic query functions look a little different, but are generally still the same:

dynamic_query_function: Callable[[PTRPCDynamicQueryFunctionMessage], Awaitable[PTRPCDynamicQueryFunctionMessageResponse]] = None,

which is to say that the function is pre-defined (one per command parameter) and looks like:

async def dynamic_query_function(myArg: PTRPCDynamicQueryFunctionMessage) -> PTRPCDynamicQueryFunctionMessageResponse:
    do something

Command OPSEC

The opsec functionality has been removed from a special CommandOPSEC class and moved to the main command class itself. So, your command can have two additional functions:

async def opsec_pre(self, taskData: PTTaskMessageAllData) -> PTTTaskOPSECPreTaskMessageResponse:
    response = PTTTaskOPSECPreTaskMessageResponse(
        TaskID=taskData.Task.ID, Success=True, OpsecPreBlocked=False,
        OpsecPreMessage="Not implemented, passing by default",
    )
    return response

async def opsec_post(self, taskData: PTTaskMessageAllData) -> PTTTaskOPSECPostTaskMessageResponse:
    response = PTTTaskOPSECPostTaskMessageResponse(
        TaskID=taskData.Task.ID, Success=True, OpsecPostBlocked=False,
        OpsecPostMessage="Not implemented, passing by default",
    )
    return response

SOCKS

The RPC call to start SOCKS is no longer control_socks. Instead, you'll use the SendMythicRPCProxyStart and SendMythicRPCProxyStop commands as detailed here.

C2 Profiles

C2 profiles also need to be updated for Mythic 3.0.0, in an extremely similar way to Payload Types.

  1. Make a directory, c2 name , in Mythic/InstalledServices

  2. Copy your entire C2 Profile's c2 name directory into InstalledServices (yes, the path will look like Mythic/InstalledServices/c2Name/c2Name)

  3. Remove c2_service.sh, mythic_service.py, and rabbitmq_config.json from your mythic folder

  4. Remove C2_RPC_Functions.py

  5. In Mythic/InstalledServices/c2Name create a main.py and a Dockerfile

  6. In your new Dockerfile, copy the contents of your old Dockerfile and change the FROM line to FROM itsafeaturemythic/mythic_python_base:latest

  7. In your mythic/c2_functions/ folder, your definition file should import mythic_container instead of mythic_c2_container (similar to what we did for agent updates).

  8. In your new main.py add:

    import mythic_container
    from [c2 name].mythic import *
    mythic_container.mythic_service.start_and_run_forever()
  9. In your c2 profile definition, add in two more attributes - server_folder_path (path to the folder where your server binary and config.json files exist), and server_binary_path (path to the binary to execute if you're doing an egress c2 profile and not a p2p profile).

To see what this looks like all together, look at the websocket example here: https://github.com/MythicMeta/ExampleContainers/tree/main/Payload_Type/python_services. You'll notice that the websocket is just one of multiple services that the single docker contianer is offering. If you want your container to only offer that one, then you can remove the other folders and adjust your main.py accordingly.

Keys in the C2 Profile Parameter Type Dictionary will be sorted alphabetically - they will NOT maintain the order they were specified in the UI. This is currently a limitation of the Golang Google JSON library.

Translation Containers

Translation containers are no different than C2 Profiles and Payload Types for the new format of things. Look to translator in the ExampleContainers (https://github.com/MythicMeta/ExampleContainers/tree/main/Payload_Type/python_services) repository for an example of how to format your new structure. Translation containers boil down to one class definition with a few functions.

One big change from Mythic 2.3 -> 3.0 for Translation Containers is that they now operate over gRPC instead of RabbitMQ. This means that they need to access the gRPC port on the Mythic Server if you intend on running a translation container on a separate host from Mythic itself. This port is configurable in the Mythic/.env file, but by default it's 17443. This change to gRPC instead of RabbitMQ for the translation container messages speeds things up and reduces the burden on RabbitMQ for transmitting potentially large messages.

Some additional notes about Translation container message updates:

  • Although Mythic 3 will base64 decode a message before providing it to translate_from_c2_format, Mythic 3 will not base64 encode the result of translate_to_c2_format which you will still need to do like you would have for Mythic 2.3

  • Mythic 2.3 allowed UUID prefixes to custom agent messages to a little endian encoded 16 byte value. In Mythic 3 any 16 byte UUID prefix needs to be big endian encoded

  • Mythic 2.3 required the translation container to base64 encode/decode inputs and outputs for generate_keys. Mythic 2.3 would directly use that base64 data to populate enc_key or dec_key values for building and would provide that base64 data directly to any translate_to_c2_format and translate_from_c2_format functions.

    • Mythic 3 expects generate_keys to provide the keys as byte arrays. Mythic 3 will base64 encode/decode the byte arrays when populating any enc_key or dec_key value for an agent configuration, but will use the byte array when calling any translate_to_c2_format and translate_from_c2_format function

  • Mythic 2.3 would provide the entire message as input to translate_from_c2_format. Mythic 3 provides the message, minus any UUID prefix

Last updated