Use GitHub's private vulnerability reporting. Please do not open a public issue for something exploitable.
Include the version and build from GET /health, the request you sent, and what
came back. If a container reproduces it, give the image tag.
One maintainer reads these. Expect an acknowledgement within a week. How long a fix takes depends on what it is.
The published image tags. Fixes go to latest and to the most recent minor
version. Older tags are not patched.
- Escaping the container, or running code in it, through a crafted URL or a crafted page.
- A parser or converter defect that lets a fetched page run script in the service's own context rather than being extracted as text.
- Reading files from the container that a request should not reach.
- One request interfering with another that is in flight.
These are how the service is meant to work:
- No authentication. Every endpoint is open to anyone who can reach the port.
The documented deployment publishes it on
127.0.0.1for that reason. - It fetches any URL it is given.
POST /processwill request an internal address, a cloud metadata endpoint or a device on your LAN. That is what a URL fetcher does. If someone who should not have that reach can also reach this service, the exposure is the network, not the fetch. - Live mode serves fetched pages from your origin.
MODE=livewrites the extracted article into the page it returns. That content comes from the internet. It is meant for one person on loopback. - Chromium runs without its sandbox, and the container runs as root. The
service passes
--no-sandboxand the container is the boundary. A page that escapes Chromium has not escaped the container. - What you are allowed to fetch is your problem. Robots.txt, terms of service and copyright are not security questions and this project cannot answer them.
Dependabot watches NuGet packages, GitHub Actions and the container base image.
NuGet advisories fail the build rather than appearing as a warning: package audit
warnings are errors via TreatWarningsAsErrors.