Skip to content

Security: selimozten/elek

SECURITY.md

Security Policy

elek is designed as a review-only GitHub Action. The model-facing MCP server exposes only create_inline_comment and update_tracking_comment; it has no code path for approving, merging, closing, or changing repository settings.

Supported Versions

Security fixes target the latest v1.x release and the moving v1 action tag. Pin selimozten/elek@v1 for normal use, or pin an exact patch tag when you need reproducible workflow behavior.

Reporting a Vulnerability

Do not open a public issue for vulnerabilities that expose credentials, repository write access, private code, or a bypass of the review-only tool surface.

Preferred reporting path:

  1. Use GitHub private vulnerability reporting for this repository if it is available.
  2. If private reporting is unavailable, open a minimal public issue asking for a private contact path. Do not include exploit details, tokens, logs, or private repository names in that issue.

Useful reports include:

  • The action version or commit SHA.
  • A minimal workflow snippet with secrets redacted.
  • The GitHub event type and permissions block.
  • Whether the run was on a fork PR.
  • The provider/model used, with API keys redacted.
  • A clear impact statement: token exposure, unauthorized write path, comment spoofing, dependency execution issue, or denial of service.

Scope

In scope:

  • Token leakage in prompts, logs, comments, MCP buffers, or temporary files.
  • A path that lets the model approve, merge, close, or mutate repository settings through elek.
  • Unsafe default permissions or tool surfaces.
  • Dependency or runtime behavior that executes untrusted code unexpectedly.
  • Comment spoofing or identity confusion that could mislead maintainers.

Out of scope:

  • Model hallucinations that do not gain privileges or leak sensitive data.
  • Expected GitHub token limitations, such as GITHUB_TOKEN reviews not satisfying required reviewer counts.
  • Findings that require the workflow owner to deliberately grant broad permissions and use mode: agent.

There aren't any published security advisories