Skip to content

Security: 200lz/llm-engineering-platform

Security

SECURITY.md

Security Policy

Supported versions

Security fixes are applied to the latest release line and the default branch. Older snapshots and local modifications may not receive updates.

Reporting a vulnerability

Please do not disclose a suspected vulnerability in a public issue, discussion, pull request, log excerpt, evaluation artifact, or dataset.

Use GitHub's private vulnerability reporting or security-advisory flow for this repository when available. If no private channel is visible, open a minimal public issue asking a maintainer to establish private contact. Do not include exploit details, credentials, private data, model content, internal URLs, or reproduction artifacts in that issue.

A useful private report includes:

  • the affected component and revision;
  • the impact and realistic threat model;
  • minimal reproduction steps;
  • whether credentials, prompts, responses, or model files are involved;
  • any suggested mitigation; and
  • whether the issue is already public elsewhere.

Maintainers will acknowledge reports on a best-effort basis, validate the impact, coordinate a fix and disclosure plan when appropriate, and credit the reporter if requested.

Security scope

The repository demonstrates production-oriented engineering but is not a hosted service. Operators remain responsible for network exposure, model provenance, host security, dependency review, data retention, and secret management.

Important documented limitations include:

  • the gateway has no authentication or rate limiting;
  • Compose ports bind to loopback by default and should not be exposed directly to an untrusted network;
  • model weights are supplied locally and are not verified or downloaded by the project;
  • the experiment registry and RAG fingerprints detect changes but are not cryptographic signatures or trusted attestation;
  • generated reports and RAG stores may contain sensitive operational metadata or bounded model output; and
  • external llama.cpp and container images have their own security lifecycle.

See docs/deployment.md for container and network assumptions and docs/reproducibility.md for artifact privacy boundaries.

There aren't any published security advisories