Skip to content

Security: christopherkarani/rv

Security

SECURITY.md

Security Policy

rv is a fail-closed policy gate. A command that should be denied but is allowed, a secret that reaches an agent, or a sandbox escape is a security bug. Please report those privately.

Reporting a vulnerability

Use GitHub's private vulnerability reporting: open the Security tab and choose Report a vulnerability. Do not open a public issue or post details in Discord.

Please include:

  • the rv version (rv --version) and platform (macOS version or Linux distro)
  • the host agent and hook involved, if any
  • the smallest command, file path, or config that shows the problem
  • what rv did and what it should have done

Supported versions

Only the latest release gets security fixes. Upgrade by re-running the installer.

Version Supported
0.2.x (latest) Yes
< 0.2 No

Scope

In scope:

  • gate bypasses: a destructive command or secret-file access that rv allows
  • fail-open behavior: errors, malformed input, or corrupt config that produce allow instead of deny or indeterminate
  • secret exposure: real secret bytes reaching an agent, a log, or the audit trail
  • isolation escapes: Seatbelt, Landlock, egress proxy, or microVM boundaries
  • forged or replayable allow-once codes, approvals, or audit entries

Out of scope:

  • commands rv is not designed to gate (see the rule packs)
  • attacks that require already running arbitrary code as your user outside any agent session
  • denial of service against your own machine by your own user

Design notes

The threat model, boundary decisions, and audit evidence live in docs/security.

There aren't any published security advisories