Skip to content

Security: skillmds/skillmd

SECURITY.md

Security Policy

Reporting a vulnerability

Email hi@skillmd.com with the details — proof of concept, affected versions, and impact as you understand it. Please do not open a public issue for anything exploitable.

You'll get an acknowledgment within a few days. We ask for a reasonable disclosure window (up to 90 days) to ship a fix before details go public, and we'll credit you in the release notes unless you prefer otherwise.

Scope

  • The npm packages skillmds and @skillmds/core (this repository)
  • The skillmds/skillmd/action GitHub Action
  • The SkillMD registry and API at skillmd.com / api.skillmd.com

What these tools already defend against

Reports that break one of these guarantees are especially valuable:

  • Installing a skill never executes code. skillmd add and the MCP skillmd_install tool only write files; scripts in a skill are surfaced as security flags, never run.
  • Path containment. Skill names are validated against a strict slug alphabet and every written file passes a zip-slip guard — nothing lands outside the target skill directory.
  • Content integrity. Registry-stored files are content-addressed; downloads are verified against their recorded SHA-256 and refused wholesale on mismatch (no partial installs).
  • Egress containment. Companion files are only ever fetched from the pinned GitHub raw hosts; any other source_url in a registry response is refused (client-side SSRF guard).

Supported versions

Only the latest published version of each package receives fixes.

There aren't any published security advisories