superuser5 / pycobalt

Python API for Cobalt Strike

PyCobalt

PyCobalt is a Python API for Cobalt Strike.

Usage

PyCobalt comes in two parts: a Python library and an Aggressor library.

Python Side

A Python script for PyCobalt looks like this:

#!/usr/bin/env python3

import pycobalt.engine as engine
import pycobalt.aggressor as aggressor
import pycobalt.aliases as aliases

# register this function as a beacon console alias
@aliases.alias('example-alias')
def example_alias(bid):
    aggressor.blog2(bid, 'example alias')

# read commands from cobaltstrike. must be called last
engine.loop()

engine.loop() tells PyCobalt to read commands from Cobalt Strike. It's only necessary if your script uses callbacks (for aliases, events, etc).

Fatal runtime exceptions and Python parser errors will show up in the script console.

PyCobalt includes the following modules:

• engine.py: Main communication code
• aggressor.py: Stubs for calling Aggressor functions
• aliases.py: Beacon console alias registration
• commands.py: Script console command registration
• events.py: Event handler registration
• gui.py: Context menu registration
• bot.py: Event Log bot toolkit
• helpers.py: Assorted helper functions and classes to make writing scripts easier
• sharpgen.py: Helper functions for using SharpGen with Cobalt Strike

Head over to the examples section for more information about each module.

Cobalt Strike Side

An Aggressor script for PyCobalt looks like this:

$pycobalt_path = '/root/tools/pycobalt/aggressor';
include($pycobalt_path . '/pycobalt.cna');
python(script_resource('my_script.py'));

It's necessary to set the $pycobalt_path variable so that PyCobalt can find its dependencies.

Installation

Python Side

Run setup.py install to install the PyCobalt python library.

Or you can run it straight out of the repo if you're familiar with PYTHONPATH.

Cobalt Strike Side

The Aggressor library is in the aggressor directory. it's also installed by setup.py at /usr/lib/python-*/site-packages/pycobalt-*/aggressor.

You can include pycobalt.cna straight out of there. It comes with its dependencies and all. See the usage section for more info.

PyCobalt depends on the org.json Java library. A copy is included in this repo at aggressor/jars/json.jar. You can optionally replace json.jar with a more trusted copy. It's PyCobalt's only binary dependency.

Examples

Here are some script examples. For more complete examples see the examples directory.

Script Console

To print a message on the script console:

import pycobalt.engine as engine

engine.message('test message')

This shows up in the script console as:

[pycobalt example.py] test message

To print an error message on the script console:

import pycobalt.engine as engine

engine.error('test error')

This shows up in the script console as:

[pycobalt example.py error] test error

To print debug messages to the script console:

import pycobalt.engine as engine

engine.enable_debug()
engine.debug('debug message 1')
engine.debug('debug message 2')
engine.disable_debug()
engine.debug('debug message 3')

This shows up in the script console as:

[pycobalt example.py debug] debug message 1
[pycobalt example.py debug] debug message 2

To print raw stuff to the script console you can just call the Aggressor print functions:

import pycobalt.engine as engine
import pycobalt.aggressor as aggressor

aggressor.println('raw message')

Aggressor

Calling an Aggressor function:

import pycobalt.engine as engine
import pycobalt.aggressor as aggressor

for beacon in aggressor.beacons():
    engine.message(beacon['user'])

Calling an Aggressor function with a callback:

import pycobalt.engine as engine
import pycobalt.aggressor as aggressor

def my_callback(bid, results):
    aggressor.blog2(bid, 'ipconfig: ' + results)

for beacon in aggressor.beacons():
    bid = beacon['bid']
    aggressor.bipconfig(bid, my_callback)

engine.loop()

Calling an Aggressor function without printing tasking information to the beacon console (! operator, only supported by certain functions):

...
aggressor.bshell(bid, 'whoami', silent=True)
...

For notes on using non-primitive objects such as dialog objects see the non-primitive objects section.

Aliases

Registering a beacon console alias:

import pycobalt.engine as engine
import pycobalt.aliases as aliases
import pycobalt.aggressor as aggressor

@aliases.alias('test_alias')
def test_alias(bid):
    aggressor.blog2(bid, 'test alias called')

engine.loop()

Registering an alias with help info:

...
@aliases.alias('test_alias', short_help='Tests alias registration')
...

By default the long help will be based on the short help and python function syntax. For example:

beacon> help test_alias
Tests alias registration

Syntax: test_alias

Or you can specify the long help yourself:

...
@aliases.alias('test_alias', 'Tests alias registration', 'Test alias\n\nLong help')
...

When the alias is called its arguments will be automagically checked against the arguments of the python function. For example:

beacon> test_alias foo
[-] Syntax: test_alias

To bypass this you can use python's * operator:

import pycobalt.engine as engine
import pycobalt.aliases as aliases
import pycobalt.aggressor as aggressor

@aliases.alias('test_alias')
def test_alias(bid, *args):
    aggressor.blog2(bid, 'test alias called with args: ' + ', '.join(args))

engine.loop()

This also allows you to use Python's argparse with aliases. For more information about using argparse see the helpers section.

If an unhandled exception occurs in your alias callback PyCobalt will catch it and print the exception information to the beacon console. For example, while I was writing the previous example I typed engine.blog2() instead of aggressor.blog2() by accident and got this error:

beacon> test_alias
[-] Caught Python exception while executing alias 'test_alias': module 'pycobalt.engine' has no attribute 'blog2'
    See Script Console for more details.

In the script console:

...
[pycobalt script error] exception: module 'pycobalt.engine' has no attribute 'blog2'
[pycobalt script error] traceback: Traceback (most recent call last):
  File "/usr/lib/python3.7/site-packages/pycobalt-1.0.0-py3.7.egg/pycobalt/engine.py", line 122, in loop
    handle_message(name, message)
  File "/usr/lib/python3.7/site-packages/pycobalt-1.0.0-py3.7.egg/pycobalt/engine.py", line 89, in handle_message
    callbacks.call(callback_name, callback_args)
  File "/usr/lib/python3.7/site-packages/pycobalt-1.0.0-py3.7.egg/pycobalt/callbacks.py", line 42, in call
    callback(*args)
  File "/usr/lib/python3.7/site-packages/pycobalt-1.0.0-py3.7.egg/pycobalt/aliases.py", line 36, in alias_callback
    raise e
  File "/usr/lib/python3.7/site-packages/pycobalt-1.0.0-py3.7.egg/pycobalt/aliases.py", line 32, in alias_callback
    callback(*args)
  File "/sandboxed/tools/cobaltstrike/scripts/recon.py", line 170, in test_alias
    engine.blog2(bid, 'test alias called with args: ' + ', '.join(args))
AttributeError: module 'pycobalt.engine' has no attribute 'blog2'

Commands

Script console commands are similar to beacon console aliases.

import pycobalt.engine as engine
import pycobalt.commands as commands

@commands.command('test_command')
def test_command():
    engine.message('test_command called')

engine.loop()

Error handling and argument checking is similar. Error messages are printed to the script console.

Events

Registering an event handler:

import pycobalt.engine as engine
import pycobalt.events as events

@events.event('beacon_initial')
def beacon_initial_handler(bid):
    aggressor.bnote(bid, 'fresh')

engine.loop()

This will raise an exception if the event isn't one of the official Cobalt Strike ones. To register an arbitrary event (e.g. for use with fireEvent):

...
@events.event('myevent', official_only=False)
...

GUI

The following menu tree pieces are supported:

• popup
• menu
• item
• insert_menu
• separator

Here's an example using all of those:

import pycobalt.engine as engine
import pycobalt.gui as gui

def beacon_top_callback(bids):
    engine.message('showing menu for: ' + ', '.format(bids))

def node_sysadmin(bids):
    for bid in bids:
        aggressor.bnote(bid, 'sysadmin!')

menu = gui.popup('beacon_top', callback=beacon_top_callback, children=[
    gui.menu('Note', children=[
        gui.insert_menu('note_top'),
        gui.item('sysadmin', callback=note_sysadmin),
        gui.separator(),
        gui.insert_menu('note_bottom'),
    ])
])
gui.register(menu)

engine.loop()

Callbacks are called before children are produced.

GUI registration must happen before engine.loop() is called. engine.loop() creates a new thread in Cobalt Strike and trying to register callbacks for menus created before that point (e.g. beacon_top) will result in a thread safety exception within Java. It's not possible to register menus using the regular Aggressor functions for the same reason.

The one downside to this is that you can't generate the menu labels dynamically from within the menu callbacks.

Bot

bot.py provides tools for registering Event Log bot commands.

For example:

import pycobalt.bot as bot
import pycobalt.engine as engine

bot.set_prefix('!')
bot.set_triggers(bot.PRIVMSG, bot.PREFIX, bot.ADDRESSED)
bot.add_help()

@bot.command('test-command', 'Tests bot')
def _(*args):
    for arg in args:
        bot.say(arg)

engine.loop()

Using the example:

event> !help test-command
10/19 10:21:01 <bot> test-command: Tests bot
Syntax: test-command(*args)

event> !test-command arg1 "arg 2" arg3
10/19 10:24:13 <bot> arg1
10/19 10:24:13 <bot> arg 2
10/19 10:24:13 <bot> arg3

See examples/bot.py for more examples.

Helpers

helpers.py contains helper functions and classes to make writing scripts easier. Here are some of the functions available:

• parse_jobs(content): Parses the output of bjobs as returned by the beacon_output_jobs event. Returns a list of dictionaries. Each dictionary represents a job with the following fields: jid (job ID), pid (process ID), and description.
• parse_ps(content): Parses the callback output of bps. Returns a list of dictionaries. Each dictionary represents a process with the following fileds: name, pid, ppid, arch (if available), and user (if available).
• parse_ls(content): Parses the callback output of bls. Returns a list of dictionaries. Each dictionary represents a file with the following fields: type (D/F), size (in bytes), modified (date and time), and name.
• recurse_ls(bid, directory, callback, depth=9999): Recursively list files with bls and call callback(path) for each file.
• find_process(bid, proc_name, callback): Calls bps to find a process by name and calls callback with a list of matching processes (as returned by parse_ps).
• explorer_stomp(bid, file): Stomps a file timestamp with the modification time of explorer.exe.
• upload_to(bid, local_file, remote_file): Like aggressor.bupload but lets you specify the remote file path/name.
• powershell_quote(arg)/pq(arg): Quote a string for use as an argument to a Powershell function. Encloses in single quotation marks with internal quotation marks escaped.
• argument_quote(arg)/aq(arg): Quote a string for use as an argument to a cmd.exe command that uses CommandLineToArgvW. Read this.
• cmd_quote(arg)/cq(arg): Quote a string for use as an arguent to a cmd.exe command that does not use CommandLineToArgvW.
• powershell_base64(string): Encode a string as UTF-16LE and base64 it. The output is compatible with Powershell's -EncodedCommand flag.

There's a helpers.ArgumentParser class which extends argparse.ArgumentParser to support printing to the beacon console, script console, or event log. Here's an example using it with an alias:

@aliases.alias('outlook', 'Retrieve an outlook folder', 'See `outlook -h`')
def _(bid, *args):
    parser = helpers.ArgumentParser(bid=bid, prog='outlook')
    parser.add_argument('-f', '--folder', help='Folder name to grab')
    parser.add_argument('-s', '--subject', help='Match subject line (glob)')
    parser.add_argument('-t', '--top', metavar='N', type=int, help='Only show top N results')
    parser.add_argument('-d', '--dump', action='store_true', help='Get full dump')
    parser.add_argument('-o', '--out', help='Output file')
    try: args = parser.parse_args(args)
    except: return
    ...

In the beacon console:

beacon> outlook -h
[-] usage: outlook [-h] [-f FOLDER] [-s SUBJECT] [-t N] [-d] [-o OUT]

optional arguments:
  -h, --help            show this help message and exit
  -f FOLDER, --folder FOLDER
                        Folder name to grab
  -s SUBJECT, --subject SUBJECT
                        Match subject line (glob)
  -t N, --top N         Only show top N results
  -d, --dump            Get full dump
  -o OUT, --out OUT     Output file

beacon> outlook -z
[-] unrecognized arguments: -z

To use helpers.ArgumentParser with the event log pass event_log=True to the constructor. This is useful for creating bots.

SharpGen

sharpgen.py provides helpers for compiling and executing C# code with SharpGen. It provides the following functions:

• compile_file(source, ...): Compile a C# file. By default this creates a temporary output file and returns its name.
• compile(code, ...): Compile inline C# code. By default this creates a temporary output file and returns its name.
• execute_file(bid, source, ...): Compile and execute a C# file.
• execute(bid, code, ...): Compile and execute inline C# code.

These functions have a number of shared keyword arguments. See the compile_file function's pydoc for the full list.

You need a compiled version of SharpGen to use this module. By default it points to the repo copy (pycobalt/third_party/SharpGen). You can use that copy but it's a Git submodule so you'll need to initialize and build it first. To do that run:

git submodule init
git submodule update
cd third_party/SharpGen
dotnet build

You can use your own copy of SharpGen by calling sharpgen.set_location('<your copy>') or by passing it on the sharpgen_location= parameter to any of the four compile/execute functions.

Here's a basic usage example:

import pycobalt.sharpgen
sharpgen.set_location('/root/tools/SharpGen')

@aliases.alias('sharpgen-exec')
def _(bid, code):
    sharpgen.execute(bid, code)

See examples/sharpgen.py for console commands and beacon aliases to go with each compile/execute function.

Non-Primitive Objects

When passed from Cobalt Strike to Python a non-primitive object's reference is stored. A string identifying this stored reference is passed to Python (let's call it a "serialized reference"). When passed back to Cobalt Strike the serialized reference is deserialized back into the original object reference.

Non-primitive objects are effectively opaque on the Python side.

This also means there's a global reference to every non-primitive object sitting around. To save memory PyCobalt allows you to remove an object's global reference after you're finished referencing it:

...
dialog = aggressor.dialog('Test dialog', {}, callback)
...
aggressor.dialog_show(dialog)
engine.delete(dialog)

I figure passing serialized references around is better than serializing entire Java objects. There's a Python library called javaobj which supports serializing and deserializing Java objects. It doesn't work well with complex Java objects though.

Sleep Functions

You can call arbitrary Sleep and Aggressor functions (including your own Aggressor functions) like this:

engine.call('printAll', [['a', 'b', 'c']])

Which turns into:

printAll(@('a', 'b', 'c'))

To call a Sleep function in its own thread without getting its return value:

engine.call('println', args=['printing from another thread'], fork=True)

You can also eval arbitrary Sleep code:

engine.eval('println("foo")')

engine.eval doesn't perform any sort of parameter marshalling or callback serialization.