Skip to content

Security: trendvidia/protoregistry

Security

SECURITY.md

Security policy

Supported versions

protoregistry is currently in the v0.x pre-stable line. Only the latest minor release receives security fixes. There is no long-term support branch yet — that will arrive with v1.0.

Version Supported
0.x (latest minor) yes
older 0.x no

Reporting a vulnerability

Please do not open a public GitHub issue for security problems.

Send a report to security@trendvidia.com with:

  • A description of the issue and the impact you observed.
  • The smallest reproducer you can share (a .proto file, a gRPC payload, the relevant CLI invocation, etc.).
  • The protoregistry version (commit SHA or tag) and Go version you tested against.
  • Whether you would like to be credited in the fix announcement.

We will acknowledge your report within 3 business days and aim to ship a fix or mitigation within 30 days for high-severity issues. We will keep you informed of progress and coordinate disclosure timing with you.

Threat model and known limitations

Operators deploying protoregistry should be aware of the following:

  • The gRPC server has no built-in authentication. Bind it only to trusted networks, or place it behind a reverse proxy / service mesh that enforces mTLS or token auth. A pluggable authentication interceptor is on the roadmap; until it lands, treat the server as administrator-only.
  • .proto compilation runs untrusted input through protocompile. Even with the input-size and timeout guards we apply, malicious schemas can consume non-trivial CPU and memory. Do not allow anonymous publishes from untrusted callers.
  • The __builtins__ namespace is privileged. Files published there are visible to every other namespace and can shadow well-known types (with --force). Restrict write access to it at the network/auth layer.
  • PostgreSQL credentials are read from --db / DATABASE_URL. Avoid passing them on the command line in shared shells; prefer the environment variable or a secrets file.

These limitations are tracked in STABILITY.md and in open issues. Reports that materially extend or contradict the list above are welcome via the disclosure channel above.

There aren't any published security advisories