Skip to content

Security: sigilco/intl-ai

Security

SECURITY.md

Security Policy

Supported versions

Only the latest minor release line receives security fixes.

Version Supported
0.7.x yes
< 0.7 no

Reporting a vulnerability

Report vulnerabilities privately through GitHub: use "Report a vulnerability" on this repository's Security tab (private vulnerability reporting). Do not open a public issue.

Reports are acknowledged as soon as possible. If a report is confirmed, a fix is prepared on a private branch and shipped in the next patch release, with credit in the release notes if you want it.

Trust boundaries

  • intl-ai runs external programs only when the config explicitly asks: kind = "command" providers, [[checks]] kind = "exec" checks, and [[formats]] exec format plugins. Exec checks and format plugins are only honored from the project-root config, never from vendored or nested configs, so pulling a dependency cannot silently run its code.
  • Secrets come from ${env:VAR} interpolation or the process environment; they are never written to locale files, the lockfile, or reports.

There aren't any published security advisories