Skip to content

Security

Skills give instructions to your assistant, and a skill can include scripts the assistant may run. A skill from a source is code you run on your own machine. Quiver serves skills from your library and from the sources you added; it does not check what they say.

  • Add only sources you trust, and read their skills before you assign them: quiver show <id> prints a skill's files and its SKILL.md.
  • Review updates. quiver update downloads new commits before it applies them, lists the changed files of every assigned skill, marks scripts and executable files as needs review, and asks before it applies the changes. Read them with quiver update --diff; answering no leaves the source as it was.
  • Pin what you depend on. quiver pin keeps a skill at a commit you reviewed. Move the pin when you have read the changes.
  • Fork to change. Source skills are read-only in Quiver. quiver fork copies one into your own skills.
  • Watch global skills. create_skill and update_skill let the assistant write skills into your library. A global skill's description reaches the assistant in every folder, so review new global skills with quiver list and quiver edit.
  • Review what a sync brings. quiver sync takes in what another machine pushed. It lists changes that make a skill always-on, change the files of an always-on skill or add or change a source, and asks before it pulls. Use a private remote.
  • Know what the whole library exposes. get_skill and search_skills read the whole library, not only the skills assigned to the project. Assign only what you want a project's sessions to see by name.
  • Always-on skills are changed by you. update_skill changes only skills served in the session's project, and never an always-on skill: every session starts with that text, so change it yourself with quiver edit.

Every change to your own skills is a commit in ~/.quiver/local (in ~/.quiver once you use quiver sync). git -C ~/.quiver/local log -p shows them, and git revert undoes one.

The session-start hook passes the text of your always-on skills and the ID and description of your other skills to the model. A skill that cannot be read is left out and named in a <quiver-notice> block. Text that reaches the model is cleaned: hidden characters are dropped, and a skill cannot write Quiver's own tags (<quiver-always-on>, <quiver-skills>, <quiver-notice>). ~/.quiver/logs/hook.log records each run with a hash of the text it passed.

A source skill folder that is a symlink is not discovered, so an update cannot change a skill through files that git reports somewhere else.

quiver install changes your assistants' config files: it adds the MCP server and the hooks. quiver uninstall removes both.

The install scripts download a release archive from GitHub, check it against the release's checksums.txt, and place one binary. Read install.sh or install.ps1 before you pipe them into a shell, or install with npm, Go or an archive instead.

npm packages are published from CI with npm trusted publishing and provenance; no npm token is stored.

Quiver makes network requests only for:

  • git operations on the sources and the sync remote you configure,
  • a daily check of the latest release on GitHub, off with QUIVER_NO_UPDATE_CHECK=1.

There is no telemetry.

Do not open a public issue. Report it privately through GitHub: open the Security tab of the repository and choose Report a vulnerability. Include the Quiver version (quiver --version), your system and the steps to reproduce. You will get an answer within a week.

Fixes go into the latest release only. The policy is in SECURITY.md.