Nostos reads the archive a person downloads from Google because they want their data back. That archive holds their photographs, their contacts, their calendar, their documents: often the most complete record of a life that exists in one place.
The application makes one promise, and the whole design rests on it:
The application process opens no network connections.
Everything else follows from that. How the promise is made executable, and how to verify it yourself, is documented in PRIVACY_AUDIT.md.
Anything that breaks one of these, whether or not it needs a malicious file to trigger:
- Data leaving the machine. Any network connection, DNS lookup or telemetry, from the Rust process or from the webview.
- Writing outside the chosen paths. The application writes only where the user pointed it. An archive entry escaping the destination (zip-slip) belongs here, and so does anything writing outside the folder a dialog returned.
- Losing data. No function deletes files. A path that destroys, truncates or overwrites something the user did not agree to lose is a vulnerability, not a bug: an export is often the only remaining copy.
- Command or argument injection. The one place the application reaches the operating system is the "Show in file manager" button, described in section 7b of the audit. Anything that turns a filename into an executed command belongs here.
- Reading a crafted file causing memory unsafety. The parsers handle untrusted input: ZIP entries, EXIF blocks, vCard, iCalendar and JSON sidecars all come from a file the user did not write.
Crashing on a malformed file is a bug, and worth reporting as an issue, but it is not a vulnerability on its own: the application processes files, so a bad file should produce an error rather than a panic.
- The system webview itself (WKWebView, WebKit2GTK, WebView2). It is an
operating system component: report those to its vendor. What is in scope is
our confinement of it, that is the Content Security Policy in
src-tauri/tauri.conf.json. - Anything requiring an attacker who already runs code as the user. At that point the export is readable without going through this application.
- The absence of code signing. It is a known gap, recorded in the changelog, not a finding.
Please do not open a public issue for a security problem. A public report tells everyone how to exploit it before there is a fix.
Use GitHub's private reporting: the Security tab of this repository, then Report a vulnerability. It opens a channel visible only to the maintainer.
Useful in a report, roughly in order of usefulness:
- what an attacker gains, in one sentence
- the steps to reproduce, and a sample file if the trigger is a crafted one
- the version, from the guide, and the operating system
- whether it needs the user to do something, and what
A file that reproduces the problem is worth more than a long description. If it holds real personal data, describe how to build an equivalent one rather than sending it: this project exists to keep such files off other people's machines, and that includes the maintainer's.
This is maintained by one person, so no response time is promised that could not be kept. What is promised instead:
- a report is acknowledged when it is read, even if the answer is that it needs time
- a confirmed finding is fixed before anything else
- the fix says what was wrong and since which version, in the changelog
- credit goes to whoever reported it, unless they prefer otherwise
Nothing has been released yet. When it is, only the latest version is supported: there is no back-porting to older ones.