Please report security issues via GitHub Security Advisories (preferred). If that is not possible, open an issue with minimal details and clearly mark it as a security concern.
Do NOT open a public GitHub issue for security vulnerabilities. Use the private advisory process so we can coordinate a fix before disclosure.
We aim to acknowledge reports within 48 hours and provide an initial assessment within 7 days.
- Description of the vulnerability
- Steps to reproduce
- Impact (what an attacker could do)
- Affected versions
| Version | Supported |
|---|---|
| Latest release | Yes |
| Previous minor | Best-effort |
| Older | No |
We use CVSS (Common Vulnerability Scoring System) to rate severity from 0 to 10. Higher scores indicate more serious vulnerabilities.
We will triage reports as quickly as possible and coordinate a fix before public disclosure.
- VoiceTerm is local-first and runs backend CLIs (
codex/claude) on the same machine as the user. - Backend subprocesses inherit the invoking user's OS privileges. VoiceTerm does not sandbox them.
- The repository/workspace you run against is part of the trust boundary. Untrusted repos are high risk.
- Network/API behavior and tool execution permissions are primarily enforced by the backend CLI being used.
The Security Guard CI lane (.github/workflows/security_guard.yml) installs
the locked cargo-audit and cargo-deny tools, runs cargo audit and
cargo deny check against the checked-in Rust dependency metadata, and runs
CodeQL analysis for the Python packaging and release helpers.