Security fixes are issued for the latest release on the main branch. Older builds are not patched; rebuild from main or update to the newest release.

Please do not open a public issue for a security problem.

Use GitHub's private reporting flow:

1. Open https://github.com/WinterGate-IC/VoiceForge/security/advisories/new.
2. Describe the issue, the affected version and, if possible, a minimal reproduction.

You will get an acknowledgement as soon as the report is reviewed. Please give us a reasonable window to ship a fix before publishing details.

In scope:

• the VoiceForge launcher, native app, browser/Flet front-ends and CLI;
• the Windows package and the offline Setup installer;
• the build and release scripts in tools/;
• the render pipeline, audio writers and library/profile parsers;
• the optional clone engine integration code.

Out of scope:

• the Microsoft Edge speech service itself (report to Microsoft);
• third-party libraries (report those upstream as well);
• issues that require an already-compromised machine or administrator rights;
• social-engineering or physical access;
• the optional XTTS v2 model weights and coqui-tts/torch installs.

• No secrets in the repository. The Edge engine needs no API key, and the release scripts never read or store a token.
• No binaries or user data are committed. .gitignore excludes virtual environments, build trees, releases, caches, downloads, voices and scripts.
• Installer hardening. Archive members are validated before extraction: absolute paths and .. traversal are rejected, and Windows extended-length paths are used so a deep install folder cannot break or escape extraction.
• Crash-safe audio. Every rendered part and the finished file is validated, written to a temporary name and renamed into place; a failed render leaves nothing half-written in the downloads folder.
• No telemetry. Scripts, voices and audio stay in the local library. Only the text you explicitly synthesize is sent to Microsoft's Edge endpoint.
• Checksums. Release assets publish a SHA-256 digest; GitHub also reports the digest of every uploaded asset.
• Dependency updates. Dependabot watches Python and GitHub Actions dependencies; CodeQL and the test suite run on every push and pull request.
• Licensing. The build ships only the RSA public key used to verify a licence. The signing key is offline in a private repository, so no public build can mint a licence. See OPEN_CORE.md.

Get-FileHash .\VoiceForge-<timestamp>.zip -Algorithm SHA256

Compare the result with the digest in the release notes. The same digest is shown on the GitHub release asset.

VoiceForge is open source under the MIT licence. There are no server-side secrets: the default engine talks directly to Microsoft's public speech endpoint, and the optional clone engine runs locally. The trade-off of MIT is that anyone may fork and redistribute the code; that is expected and is separate from vulnerability handling.