Commands

What do Commands track?

Command information is tracked in your Payload Type's container. Each command has its own Python class or GoLang struct. In Python, you leverage CommandBase and TaskArguments to define information about the command and information about the command's arguments.

CommandBase defines the metadata about the command as well as any pre-processing functionality that takes place before the final command is ready for the agent to process. This class includes the create_go_tasking (Create_Tasking) and process_response (Process Response) functions.

****TaskArguments does two things:

1. defines the parameters that the command needs

2. verifies / parses out the user supplied arguments into their proper components

   • this includes taking user supplied free-form input (like arguments to a sleep command - 10 4) and parsing it into well-defined JSON that's easier for the agent to handle (like {"interval": 10, "jitter": 4}). This can also take user-supplied dictionary input and parse it out into the rightful CommandParameter objects.

   • This also includes verifying all the necessary pieces are present. Maybe your command requires a source and destination, but the user only supplied a source. This is where that would be determined and error out for the user. This prevents you from requiring your agent to do that sort of parsing in the agent.

If you're curious how this all plays out in a diagram, you can find one here: #operator-submits-tasking.

CommandBase

from mythic_payloadtype_container.PayloadBuilder import *
from mythic_payloadtype_container.MythicCommandBase import *

class ScreenshotCommand(CommandBase):
    cmd = "screenshot"
    needs_admin = False
    help_cmd = "screenshot"
    description = "Use the built-in CGDisplay API calls to capture the display and send it back over the C2 channel. No need to specify any parameters as the current time will be used as the file name"
    version = 1
    author = ""
    attackmapping = ["T1113"]
    argument_class = ScreenshotArguments
    browser_script = BrowserScript(script_name="screenshot", author="@its_a_feature_")
    attributes = CommandAttributes(
        spawn_and_injectable=True,
        supported_os=[SupportedOS.MacOS],
        builtin=False,
        load_only=False,
        suggested_command=False,
    )
    script_only = False
    
    async def create_go_tasking(self, taskData: MythicCommandBase.PTTaskMessageAllData) -> MythicCommandBase.PTTaskCreateTaskingMessageResponse:
        response = MythicCommandBase.PTTaskCreateTaskingMessageResponse(
            TaskID=taskData.Task.ID,
            Success=True,
        )
        return response
        
    async def process_response(self, task: PTTaskMessageAllData, response: any) -> PTTaskProcessResponseMessageResponse:
        resp = PTTaskProcessResponseMessageResponse(TaskID=task.Task.ID, Success=True)
        return resp

Creating your own command requires extending this CommandBase class (i.e. class ScreenshotCommand(CommandBase) and providing values for all of the above components.

• cmd - this is the command name. The name of the class doesn't matter, it's this value that's used to look up the right command at tasking time

• needs_admin - this is a boolean indicator for if this command requires admin permissions

• help_cmd - this is the help information presented to the user if they type help [command name] from the main active callbacks page

• description - this is the description of the command. This is also presented to the user when they type help.

• suported_ui_features - This is an array of values that indicates where this command might be used within the UI. For example, from the active callbacks page, you see a table of all the callbacks. As part of this, there's a dropdown you can use to automatically issue an exit task to the callback. How does Mythic know which command to actually send? It's this array that dictates that. The following are used by the callback table, file browser, and process listing, but you're able to add in any that you want and leverage them via browser scripts for additional tasking:

  • supported_ui_features = ["callback_table:exit"]

  • supported_ui_features = ["file_browser:list"]

  • supported_ui_features = ["process_browser:list"]

  • supported_ui_features = ["file_browser:download"]

  • supported_ui_features = ["file_browser:remove"]

  • supported_ui_features = ["file_browser:upload"]

• version - this is the version of the command you're creating/editing. This allows a helpful way to make sure your commands are up to date and tracking changes

• argument_class - this correlates this command to a specific TaskArguments class for processing/validating arguments

• attackmapping - this is a list of strings to indicate MITRE ATT&CK mappings. These are in "T1113" format.

• agent_code_path is automatically populated for you like in building the payload. This allows you to access code files from within commands in case you need to access files, functions, or create new pieces of payloads. This is really useful for a load command so that you can find and read the functions you're wanting to load in.

• You can optionally add in the attributes variable. This is a new class called CommandAttributes where you can set whether or not your command supports being injected into a new process (some commands like cd or exit don't make sense for example). You can also provide a list of supported operating systems. This is helpful when you have a payload type that might compile into multiple different operating system types, but not all the commands work for all the possible operating systems. Instead of having to write "not implemented" or "not supported" function stubs, this will allow you to completely filter this capability out of the UI so users don't even see it as an option.

  • Available options are:

    • supported_os an array of SupportedOS fields (ex: [SupportedOS.MacOS])

    • spawn_and_injectable is a boolean to indicate if the command can be injected into another process

    • builtin is a boolean to indicate if the command should be always included in the build process and can't be unselected

    • load_only is a boolean to indicate if the command can't be built in at the time of payload creation, but can be loaded in later

    • suggested_command is a boolean to indicate if the command should be pre-selected for users when building a payload

    • filter_by_build_parameter is a dictionary of parameter_name:value for what's required of the agent's build parameters. This is useful for when some commands are only available depending on certain values when building your agent (such as agent version).

    • You can also add in any other values you want for your own processing. These are simply key=value pairs of data that are stored. Some people use this to identify if a command has a dependency on another command. This data can be fetched via RPC calls for things like a load command to see what additional commands might need to be included.

  • This ties into the CommandParameter fields choice_filter_by_command_attributes, choices_are_all_commands, and choices_are_loaded_commands.

• The create_go_tasking function is very broad and covered in Create_Tasking

• The process_response is similar, but allows you to specify that data shouldn't automatically be processed by Mythic when an agent checks in, but instead should be passed to this function for further processing and to use Mythic's RPC functionality to register the results into the system. The data passed here comes from the post_response message (Process Response).

• The script_only flag indicates if this Command will be use strictly for things like issuing subtasking, but will NOT be compiled into the agent. The nice thing here is that you can now generate commands that don't need to be compiled into the agent for you to execute. These tasks never enter the "submitted" stage for an agent to pick up - instead they simply go into the create_tasking scenario (complete with subtasks and full RPC functionality) and then go into a completed state.

TaskArguments

The TaskArguments class defines the arguments for a command and defines how to parse the user supplied string so that we can verify that all required arguments are supplied. Mythic now tracks where tasking came from and can automatically handle certain instances for you. Mythic now tracks a tasking_location field which has the following values:

• command_line - this means that the input you're getting is just a raw string, like before. It could be something like x86 13983 200 with a series of positional parameters for a command, it could be {"command": "whoami"} as a JSON string version of a dictionary of arguments, or anything else. In this case, Mythic really doesn't know enough about the source of the tasking or the contents of the tasking to provide more context.

• parsed_cli - this means that the input you're getting is a dictionary that was parsed by the new web interface's CLI parser. This is what happens when you type something on the command line for a command that has arguments (ex: shell whoami or shell -command whoami). Mythic can successfully parse out the parameters you've given into a single parameter_group and gives you a dictionary of data.

• modal - this means that the input you're getting is a dictionary that came from the tasking modal. Nothing crazy here, but it does at least mean that there shouldn't be any silly shenanigans with potential parsing issues.

• browserscript - if you click a tasking button from a browserscript table and that tasking button provides a dictionary to Mythic, then Mythic can forward that down as a dictionary. If the tasking button from a browserscript table submits a String instead, then that gets treated as command_line in terms of parsing.

With this ability to track where tasking is coming from and what form it's in, an agent's command file can choose to parse this data differently. By default, all commands must supply a parse_arguments function in their associated TaskArguments subclass. If you do nothing else, then all of these various forms will get passed to that function as strings (if it's a dictionary it'll get converted into a JSON string). However, you can provide another function, parse_dictionary that can handle specifically the cases of parsing a given dictionary into the right CommandParameter objects as shown below:

async def parse_arguments(self):
    if len(self.command_line) == 0:
        raise ValueError("Must supply arguments")
    raise ValueError("Must supply named arguments or use the modal")

async def parse_dictionary(self, dictionary_arguments):
    self.load_args_from_dictionary(dictionary_arguments)

In self.args we define an array of our arguments and what they should be along with default values if none were provided.

In parse_arguments we parse the user supplied self.command_line into the appropriate arguments. The hard part comes when you allow the user to type arguments free-form and then must parse them out into the appropriate pieces.

class LsArguments(TaskArguments):
    def __init__(self, command_line, **kwargs):
        super().__init__(command_line, **kwargs)
        self.args = [
            CommandParameter(
                name="path",
                type=ParameterType.String,
                default_value=".",
                description="Path of file or folder on the current system to list",
                parameter_group_info=[ParameterGroupInfo(
                    required=False
                )]
            )
        ]

    async def parse_arguments(self):
        self.add_arg("path", self.command_line)

    async def parse_dictionary(self, dictionary):
        if "host" in dictionary:
            # then this came from the file browser
            self.add_arg("path", dictionary["path"] + "/" + dictionary["file"])
            self.add_arg("file_browser", type=ParameterType.Boolean, value=True)
        else:
            self.load_args_from_dictionary(dictionary)

The main purpose of the TaskArguments class is to manage arguments for a command. It handles parsing the command_line string into CommandParameters, defining the CommandParameters, and providing an easy interface into updating/accessing/adding/removing arguments as needed.

As part of the TaskArguments subclass, you have access to the following pieces of information:

• self.command_line - the parameters sent down for you to parse

• self.raw_command_line - the original parameters that the user typed out. This is useful in case you have additional pieces of information to process or don't want information processed into the standard JSON/Dictionary format that Mythic uses.

• self.tasking_location - this indicates where the tasking came from

• self.task_dictionary - this is a dictionary representation of the task you're parsing the arguments for. You can see things like the initial parameter_group_name that Mythic parsed for this task, the user that issued the task, and more.

• self.parameter_group_name - this allows you to manually specify what the parameter group name should be. Maybe you don't want Mythic to do automatic parsing to determine the parameter group name, maybe you have additional pieces of data you're using to determine the group, or maybe you plan on adjusting it alter on. Whatever the case might be, if you set self.parameter_group_name = "value", then Mythic won't continue trying to identify the parameter group based on the current parameters with values.

The class must implement the parse_arguments method and define the args array (it can be empty). This parse_arguments method is the one that allows users to supply "short hand" tasking and still parse out the parameters into the required JSON structured input. If you have defined command parameters though, the user can supply the required parameters on the command line (via -commandParameterName or via the popup tasking modal via shift+enter).

When syncing the command with the UI, Mythic goes through each class that extends the CommandBase, looks at the associated argument_class, and parses that class's args array of CommandParameters to create the pop-up in the UI. While the TaskArgument's parse_arguments method simply parses the user supplied input into JSON, it's the CommandParameter's class that actually verifies that every required parameter has a value, that all the values are appropriate, and that default values are supplied if necessary.

CommandParameters

CommandParameters, similar to BuildParameters, provide information for the user via the UI and validates that the values are all supplied and appropriate.

class CommandParameter:
    def __init__(
        self,
        name: str,
        type: ParameterType,
        display_name: str = None,
        cli_name: str = None,
        description: str = "",
        choices: [any] = None,
        default_value: any = None,
        validation_func: callable = None,
        value: any = None,
        supported_agents: [str] = None,
        supported_agent_build_parameters: dict = None,
        choice_filter_by_command_attributes: dict = None,
        choices_are_all_commands: bool = False,
        choices_are_loaded_commands: bool = False,
        dynamic_query_function: callable = None,
        parameter_group_info: [ParameterGroupInfo] = None
    ):
        self.name = name
        if display_name is None:
            self.display_name = name
        else:
            self.display_name = display_name
        if cli_name is None:
            self.cli_name = name 
        else:
            self.cli_name = cli_name
        self.type = type
        self.user_supplied = False # keep track of if this is using the default value or not
        self.description = description
        if choices is None:
            self.choices = []
        else:
            self.choices = choices
        self.validation_func = validation_func
        if value is None:
            self._value = default_value
        else:
            self.value = value
        self.default_value = default_value
        self.supported_agents = supported_agents if supported_agents is not None else []
        self.supported_agent_build_parameters = supported_agent_build_parameters if supported_agent_build_parameters is not None else {}
        self.choice_filter_by_command_attributes = choice_filter_by_command_attributes if choice_filter_by_command_attributes is not None else {}
        self.choices_are_all_commands = choices_are_all_commands
        self.choices_are_loaded_commands = choices_are_loaded_commands
        self.dynamic_query_function = dynamic_query_function
        if not callable(dynamic_query_function) and dynamic_query_function is not None:
            raise Exception("dynamic_query_function is not callable")
        self.parameter_group_info = parameter_group_info
        if self.parameter_group_info is None:
            self.parameter_group_info = [ParameterGroupInfo()]

• name - the name of the parameter that your agent will use. cli_name is an optional variation that you want user's to type when typing out commands on the command line, and display_name is yet another optional name to use when displaying the parameter in a popup tasking modal.

• type- this is the parameter type. The valid types are:

  • String

  • Boolean

  • File

    • Upload a file through your browser. In your create tasking though, you get a String UUID of the file that can be used via SendMythicRPC* calls to get more information about the file or the file contents

  • Array

    • An Array of string values

  • ChooseOne

  • ChooseMultiple

    • An Array of string values

  • Credential_JSON

    • Select a specific credential that's registered in the Mythic credential store. In your create tasking, get a JSON representation of all data for a credential

  • Number

  • Payload

    • Select a payload that's already been generated and get the UUID for it. This is helpful for using that payload as a template to automatically generate another version of it to use as part of lateral movement or spawning new agents.

  • ConnectionInfo

    • Select the Host, Payload/Callback, and P2P profile for an agent or callback that you want to link to via a P2P mechanism. This allows you to generate random parameters for payloads (such as named-pipe names) and not require you to remember them when linking. You can simply select them and get all of that data passed to the agent.

    • When this is up in the UI, you can also track new payloads on hosts in case Mythic isn't aware of them (maybe you moved and executed payloads in a method outside of Mythic). This allows Mythic to track that payload X is now on host Y and you can use the same selection process as the first bullet to filter down and select it for linking.

  • LinkInfo

    • Get a list of all active/dead P2P connections for a given agent. Selecting one of these links gives you all the same information that you'd get from the ConnectionInfo parameter. The goal here is to allow you to easily select to "unlink" from an agent or to re-link to a very specific agent on a host that you were previously connected to.

• description - this is the description of the parameter that's presented to the user when the modal pops up for tasking

• choices - this is an array of choices if the type is ChooseOne or ChooseMultiple

  • If your command needs you to pick from the set of commands (rather than a static set of values), then there are a few other components that come into play. If you want the user to be able to select any command for this payload type, then set choices_are_all_commands to True. Alternatively, you could specify that you only want the user to choose from commands that are already loaded into the callback, then you'd set choices_are_loaded_commands to True. As a modifier to either of these, you can set choice_filter_by_command_attributes to filter down the options presented to the user even more based on the parameters of the Command's attributes parameter. This would allow you to limit the user's list down to commands that are loaded into the current callback that support MacOS for example. An example of this would be:

CommandParameter(name="test name", 
                 type=ParameterType.ChooseMultiple, 
                 description="so many choices!", 
                 choices_are_all_commands=True,
                 choice_filter_by_command_attributes={"supported_os": [SupportedOS.MacOS]}),

• validation_func - this is an additional function you can supply to do additional checks on values to make sure they're valid for the command. If a value isn't valid, an exception should be raised

• value - this is the final value for the parameter; it'll either be the default_value or the value supplied by the user

• default_value - this is a value that'll be set if the user doesn't supply a value

• supported_agents - If your parameter type is Payload then you're expecting to choose from a list of already created payloads so that you can generate a new one. The supported_agents list allows you to narrow down that dropdown field for the user. For example, if you only want to see agents related to the apfell payload type in the dropdown for this parameter of your command, then set supported_agents=["apfell"] when declaring the parameter.

• supported_agent_build_parameters - allows you to get a bit more granular in specifying which agents you want to show up when you select the Payload parameter type. It might be the case that a command doesn't just need instance of the atlas payload type, but maybe it only works with the Atlas payload type when it's compiled into .NET 3.5. This parameter value could then be supported_agent_build_parameters={"atlas": {"version":"3.5"}} . This value is a dictionary where the key is the name of the payload type and the value is a dictionary of what you want the build parameters to be.

• dynamic_query_function - More information can be found here, but you can provide a function here for ONLY parameters of type ChooseOne or ChooseMultiple where you dynamically generate the array of choices you want to provide the user when they try to issue a task of this type.

Most command parameters are pretty straight forward - the one that's a bit unique is the File type (where a user is uploading a file as part of the tasking). When you're doing your tasking, this value will be the base64 string of the file uploaded.

parameter_group_info

To help with conditional parameters, Mythic 2.3 introduced parameter groups. Every parameter must belong to at least one parameter group (if one isn't specified by you, then Mythic will add it to the Default group and make the parameter required).

You can specify this information via the parameter_group_info attribute on CommandParameter class. This attribute takes an array of ParameterGroupInfo objects. Each one of these objects has three attributes: group_name (string), required(boolean) ui_position (integer). These things together allow you to provide conditional parameter groups to a command.

Let's look at an example - the new apfell agent's upload command now leverages conditional parameters. This command allows you to either:

• specify a remote_path and a filename - Mythic then looks up the filename to see if it's already been uploaded to Mythic before. If it has, Mythic can simply use the same file identifier and pass that along to the agent.

• specify a remote_path and a file - This is uploading a new file, registering it within Mythic, and then passing along that new file identifier

Notice how both options require the remote_path parameter, but the file and filename parameters are mutually exclusive.

class UploadArguments(TaskArguments):
    def __init__(self, command_line, **kwargs):
        super().__init__(command_line, **kwargs)
        self.args = [
            CommandParameter(
                name="file", cli_name="new-file", display_name="File to upload", type=ParameterType.File, description="Select new file to upload",
                parameter_group_info=[
                    ParameterGroupInfo(
                        required=True,
                        group_name="Default"
                    )
                ]
            ),
            CommandParameter(
                name="filename", cli_name="registered-filename", display_name="Filename within Mythic", description="Supply existing filename in Mythic to upload",
                type=ParameterType.ChooseOne,
                dynamic_query_function=self.get_files,
                parameter_group_info=[
                    ParameterGroupInfo(
                        required=True,
                        group_name="specify already uploaded file by name"
                    )
                ]
            ),
            CommandParameter(
                name="remote_path",
                cli_name="remote_path",
                display_name="Upload path (with filename)",
                type=ParameterType.String,
                description="Provide the path where the file will go (include new filename as well)",
                parameter_group_info=[
                    ParameterGroupInfo(
                        required=True,
                        group_name="Default",
                        ui_position=1
                    ),
                    ParameterGroupInfo(
                        required=True,
                        group_name="specify already uploaded file by name",
                        ui_position=1
                    )
                ]
            ),
        ]

So, the file parameter has one ParameterGroupInfo that calls out the parameter as required. The filename parameter also has one ParameterGroupInfo that calls out the parameter as required. It also has a dynamic_query_function that allows the task modal to run a function to populate the selection box. Lastly, the remote_path parameter has TWO ParameterGroupInfo objects in its array - one for each group. This is because the remote_path parameter applies to both groups. You can also see that we have a ui_position specified for these which means that regardless of which option you're viewing in the tasking modal, the parameter remote_path will be the first parameter shown. This helps make things a bit more consistent for the user.

If you're curious, the function used to get the list of files for the user to select is here:

async def get_files(self, inputMsg: PTRPCDynamicQueryFunctionMessage) -> PTRPCDynamicQueryFunctionMessageResponse:
        fileResponse = PTRPCDynamicQueryFunctionMessageResponse(Success=False)
        file_resp = await MythicRPC().execute("get_file", callback_id=inputMsg.Callback,
                                              limit_by_callback=False,
                                              filename="",
                                              max_results=-1)
        if file_resp.status == MythicRPCStatus.Success:
            file_names = []
            for f in file_resp.response:
                if f["filename"] not in file_names and f["filename"].endswith(".exe"):
                    file_names.append(f["filename"])
            fileResponse.Success = True
            fileResponse.Choices = file_names
            return fileResponse
        else:
            fileResponse.Error = file_resp.error
            return fileResponse

In the above code block, we're searching for files, not getting their contents, not limiting ourselves to just what's been uploaded to the callback we're tasking, and looking for all files (really it's all files that have "" in the name, which would be all of them). We then go through to de-dupe the filenames and return that list to the user.

Processing Order

So, with all that's going on, it's helpful to know what gets called, when, and what you can do about it.