Skip to content

Security: Unity-Technologies/uvcs-agent-plugin

SECURITY.md

Security policy

Reporting a vulnerability

Do not open a public GitHub issue for security problems.

Report vulnerabilities to Unity's security team following the process at https://unity.com/security (or use GitHub's private security advisory form on this repository).

Please include what you found, how to reproduce it, and the impact you believe it has. We will acknowledge your report and keep you informed as we investigate.

Scope

This plugin ships documentation and skills that instruct an AI agent to run cm commands. The security-relevant surface is therefore narrow but real:

  • Instructions that cause destructive or unintended UVCS operations — for example a skill that leads an agent to check in, merge, undo, or delete without user confirmation. Treat this as a security issue, not just a bug.
  • Command injection or unsafe quoting in documented commands or in skills/*/scripts/, especially where branch names, comments, or paths are interpolated.
  • Secret or data leakage — a skill that encourages sending repository contents, credentials, tokens, server hostnames, or user data to a third party, or that instructs an agent to read credential files.
  • Prompt-injection footguns — instructions that make an agent treat untrusted repository content (changeset comments, file contents, branch attributes) as commands to obey.

Vulnerabilities in the UVCS product itself, in the cm CLI, or in a UVCS server are out of scope for this repository — report those through https://unity.com/security as Unity Version Control issues.

For users of this plugin

  • Skills here are written to keep read operations free and to require explicit confirmation before anything that mutates a repository. If an agent using this plugin performs a mutating operation without asking, please report it.
  • Review what your agent is about to run. cm acts with your credentials and your permissions.

There aren't any published security advisories