Skip to main content

Sharing

Publishing is a deliberate, separate step from using a STAK locally. This page is the precise boundary of what leaves your machine when you run stak publish.

Public STAKs

A public STAK is a named, versioned bundle published as @owner/name. Every Skill in the bundle publishes the same way: as an immutable, content-addressed snapshot pinned by content hash. That holds whether or not the Skill has a git-tracked source — a consumer never needs the original repository to exist to use a published STAK. When a Skill does have a git-tracked source, its remote/revision/path are preserved as display and update-check metadata, but they are never required for restoration. The only requirement to publish is that you've run stak push for that content first; otherwise publishing fails with an actionable error. This applies equally when publishing or sharing a Skill from the web — if it hasn't been uploaded to STAK Cloud yet, run stak push first. Anyone can stak use or stak try a published STAK; using it never grants write access back to your account or machine.

stak publish ./my-stak --slug my-stak

Portable Skill snapshots

What actually gets published is a portable snapshot of the Skills in that STAK — the instructions, context, and tooling definitions themselves, stripped of anything specific to your machine or account. That's the entire payload of a public STAK.

Versioning

Each publish is a distinct version. There is no automatic update mechanism: an already-active public STAK on someone's machine is never silently upgraded to a newer version. Consumers re-run stak use explicitly when they want to pick up changes.

What gets shared

  • The STAK's name and published slug (@owner/name).
  • The portable Skill snapshots that make up that STAK's version — instructions, context, and tool definitions.

What never gets shared

Public STAKs contain only portable Skill snapshots. They never include:

  • Memory — your agent's private, Cloud-backed context is never part of a published STAK.
  • Projects — logical Project identity and the Memory behind a Project are never published.
  • Machine state — nothing about which machine a Skill was authored or used on travels with it.
  • Credentials — tokens, API keys, and auth state are never read, stored in, or published with a STAK.
  • Local filesystem paths — directory locations on your machine, including local Project mappings, are never included.

See Memory & Projects for how Memory and Project identity work, and Security & Privacy for the full privacy model.