Gato Extreme Edition is a hard fork of Gato, which was originally developed by @AdnaneKhan, @mas0nd, and @DS-koolaid. Gato-X is maintained by @AdnaneKhan and serves to automate advanced enumeration and exploitation techniques against GitHub repositories and organizations for security research purposes.
Gato-X accompanies the BlackHat USA 2024 talk: Self-Hosted GitHub CI/CD Runners: Continuous Integration, Continuous Destruction and the DEF CON 32 talk Grand Theft Actions: Abusing Self-Hosted GitHub Runners at scale.
Gato-X is a powerful tool and should only be used for ethical security research purposes.
If scanning repositories, then the search and enumerate modes are
safe to run. They will perform no mutable operations and only perform read requests.
As of writing, enumeration will not generate any audit log events.
Gato-X automates the "Runner-on-Runner" (RoR) technique used extensively by Adnan Khan and John Stawinski during their self-hosted runner bug bounty campaign. This feature replaces the basic attack PoC functionality included in the original version of Gato.
Gato-X supports deploying RoR through fork pull requests. Gato-X also supports
creating a RoR payload only, which can be used in conjunction with the push workflow
functionality to jump to internal self-hosted runners.
Under the hood, Gato-X will perform the following steps:
- Prepare RoR C2 Repository
- Prepare payload Gist files
- Deploy the RoR implantation payload.
- Confirm successful callback and runner installation.
- Provide user with an interactive webshell upon successful connection.
From the user's persective, it's simply: run command, get shell. What more could a hacker want?
Gato-X contains a powerful scanning engine for GitHub Actions Injection and Pwn Request vulnerabilities. As of writing, Gato-X is one of the fastest tools for the task. It is capable of scanning 35-40 thousand repositories in 1-2 hours using a single GitHub PAT. This is the most sophisticated new feature in Gato-X and is the result of countless hours of development and iteration in my spare time over the last six months.
- Reachability Analysis
- Same and Cross-Repository Transitive Workflow Analysis
- Parsing and Simulation of "If Statements"
- Gate Check Detection (permission checks, etc.)
- Lightweight Source-Sink Analysis for Variables
- Priority Guidelines
As an operator facing tool, Gato-X is tuned with a higher false positive rate than a tool designed to generate alerts, but it provides contextual information to quickly determine if something is worth investigating or not. To aid in triage, Gato-X attempts to apply confidence ratings to its reports.
- Improved Secrets Exfiltration.
- Enumeration of deployment environment secrets.
- Speed improvements for runlog analysis.
- General speed improvements throughout.
- Improved CLI interface and reports.
- Removed dependancy on Git.
There are a number of features I plan to add to Gato-X in the coming weeks or months.
Currently, Gato-X does not analyze referenced composite actions. In some cases, risky operations can be performed within composite actions (such as referencing user-controlled context variables or checking out PR code).
The problem with this is that retrieving an additional file requires an addition round trip request. This can significantly slow down enumeration. This will probably be an option that is disabled by default when I add it.
Gato-X's biggest weakness is identifying injection points that are outputs of steps that run arbitrary code. This creates a lot of 'UNKNOWN' confidence Actions Injection Reports. Using LLMs to reason about whether a variable is user control or not based on context will allow further narrowing down results. This feature will likely include support for passing an OpenAI API key and some Gato-X system prompts that I will use to inform ChatGPT of what to look for and how to respond.
Similarly, this can apply to capture code that performs permission checks or enforces immutable references.
In order to motivate hackers to use Gato-X to find and report vulnerabilities in open-source repositories, I've created a mini hall of fame. If you found an issue with Gato-X and reported it, feel free to follow the instructions on the page so I can add your accomplishment for all to see!
To perform a public repository self-hosted runner takeover attack, Gato-X requires a PAT with the following scopes:
repo, workflow, and gist.
This should be a PAT for an account that is a contributor to the target repository (i.e. submitted a typo fix).
gato-x a --runner-on-runner --target ORG/REPO --target-os [linux,osx,windows] --target-arch [arm,arm64,x64]
It is very rare that maintainers select allowing workflows on pull request from all external users without approval, but it has happened.
Next, Gato-X will automatically prepare a C2 repository and begin the operation. Gato-X will monitor each step as the attack continues, exiting as gracefully as possible at each phase in case of a failure. If workflow approval is required, Gato-X will wait a short period of time before exiting.
If the full chain succeeds, Gato-X will drop to an interactive prompt. This will execute shell commands on the self-hosted runner.
If the target runner is non-ephemeral, use the --keep-alive flag. This will keep the workflow running. GitHub
Actions allows workflow runs on self-hosted runners to run for up to 5 days (as of writing, this might change - it was 30 days).
First, create a GitHub PAT with the repo scope. Set that PAT to the
GH_TOKEN environment variable.
Next, use the search feature to retrieve a list of candidate repositories:
gato-x s -sg -q 'count:75000 /(issue_comment|pull_request_target|issues:)/ file:.github/workflows/ lang:yaml' -oT checks.txt
Finally, run Gato-X on the list of repositories:
gato-x e -R checks.txt -sr | tee gatox_output.txt
This will take some time depending on your computer and internet connection speed. Since the results are very long, use tee to save them to a file
for later review. Gato-X also supports JSON output, but that is intended for further machine analysis.
These automated attacks only scratch the surface of the kinds of post-compromise attacks paths that a red teamer may encounter within large GitHub Enterprise tenants. See the wiki for complex cases and how Gato-X may help.
- Deploying RoR using custom workflow via the push trigger.
- Deploying RoR using a PAT that only has the
reposcope but can obtain execution viaworkflow_dispatch/pushtriggers. - Leveraging a
reposcoped token to bypass external contributor approval requirements, but leveraging Gato-X for RoR infrastructure setup. - Using a
GITHUB_TOKENwithactions: writefrom a Pwn Request to approve a fork PR from an external contributor.
Gato supports OS X and Linux with at least Python 3.10.
Gato-X is published on PyPi, so you can simply install it with pip install gato-x
In order to install the tool from source, simply clone the repository and use pip install.
We recommend performing this within a virtual environment.
git clone https://github.com/AdnaneKhan/gato-x
cd gato-x
python3 -m venv venv
source venv/bin/activate
pip install .
If you need to make on-the-fly modifications, then install it in editable mode with pip install -e.
After installing the tool, it can be launched by running gato-x.
We recommend viewing the parameters for the base tool using gato -h, and the
parameters for each of the tool's modules by running the following:
gato-x search -hgato-x enum -hgato-x attack -h
The tool requires a GitHub classic PAT in order to function. To create one, log
in to GitHub, go to GitHub Developer Settings
and select Generate New Token and then Generate new token (classic).
After creating this token set the GH_TOKEN environment variable within your
shell by running export GH_TOKEN=<YOUR_CREATED_TOKEN>. Alternatively, enter it when the application
prompts you.
For troubleshooting and additional details, such as installing in developer mode or running unit tests, please see the wiki.
As an operator facing tool with rapidly developed features, Gato-X will have bugs. Typically, these are related to edge cases with run log formatting or YAML files.
If you believe you have identified a bug within the software, please open an issue containing the tool's output, along with the actions you were trying to conduct.
Contributions are welcome! Please review the design methodology before working on a new feature!
Additionally, if you are proposing significant changes to the tool, please open an issue open an issue to start a conversation about the motivation for the changes.
Gato-X is licensed under the Apache License, Version 2.0.
Gato-X:
Copyright 2024, Adnan Khan
Original Gato Implementation:
Copyright 2023 Praetorian Security, Inc
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.
