Skip to content

Security: snpNEXT/Marinara-Agents

Security

SECURITY.md

Security Policy

Reporting a vulnerability

Do not open a public issue for a vulnerability that could expose user data or execute code. Use the repository's Security tab to submit a private vulnerability report. Include the affected package or publishing path, the Engine version, reproduction steps, and the impact.

Package trust model

Marinara Agent packages may contain server and client entrypoints. Server entrypoints run inside the Marinara Engine process; manifest permissions describe intended access but are not an operating-system sandbox. Treat every package, artifact, catalog entry, build script, and publishing workflow as executable supply-chain material.

Catalog SHA-256 values detect accidental corruption and mismatched downloads. They do not protect against an attacker who can change both an artifact and its catalog hash. The repository therefore also:

  • requires canonical same-repository artifact URLs during catalog validation;
  • maps repository paths to owner SpicyMarinara in CODEOWNERS;
  • validates source manifests, archive contents, file hashes, catalog lanes, and generated outputs in pull requests; and
  • pins third-party GitHub Actions by full commit SHA.

Repository administrators should keep staging protected with pull requests, catalog validation, CodeRabbit, review-thread resolution, and the required owner-approval status check. That check must exempt only active Pasta-Devs organization members and owners; every outside contributor needs a current-head approval from SpicyMarinara, and membership or API failures must fail closed. Do not add a repository-role or team bypass that could admit outside collaborators or exclude private organization members. Keep force pushes and branch deletion disabled. Protect main separately so only SpicyMarinara may promote tested work from staging.

There aren't any published security advisories