Please report security issues privately through GitHub: Security → Report a vulnerability on this repository. Do not open a public issue for anything exploitable.
Include what you found, how to reproduce it, and the version or commit. This is a personal project maintained in spare time, so expect a reply within a couple of weeks, not hours.
Only the latest commit on main is supported.
ZeroG FileMan is a local Electron app. It works on your own files with your own user permissions. These rules hold:
- Renderer isolation. The page talks to the app only through the
window.filemanpreload bridge. The main process validates every path it is given (resolveLocalPath) before touching the filesystem, and the window refuses to navigate away from the app's own origin. - No shell interpolation. External programs, such as a terminal opened at a folder, are spawned with argument arrays, never through a shell string.
- Secrets stay in main. The optional speech server API key is encrypted with Electron
safeStorage. The renderer can save, clear or check the key, but cannot read it back. If OS encryption is unavailable, the key is held in memory only and never written to disk. - Narrow network surface. The page CSP allows connections only to the app itself and to the Hugging Face hosts that the built-in Whisper engine downloads its model from (once, then cached). Fonts are bundled, so launching the app makes no network requests. Requests to a user-configured speech server are made by the main process (see ADR-0005).
| Speech engine | Audio leaves the machine? |
|---|---|
| Demo (typed transcript) | No |
| Built-in Whisper | No. Only the model download contacts Hugging Face. |
| OpenAI-compatible server | Yes, to the URL you configure, along with your API key if one is set. The settings panel warns when that URL is not on this machine, and in red when a key would be sent over plain http. |
There is no telemetry. Other than the Whisper model download and your configured speech server, the app makes no network requests.
- Anything that requires an attacker who already runs code as your user.
- Unsigned Windows builds triggering SmartScreen (see the README).