Skip to content

Security: richhaase/toolshed

Security

SECURITY.md

Security

Toolshed distributes Agent Skills plugins to Claude Code and Codex. Skills ship as Markdown plus optional Bash and Node.js scripts that execute on the user's machine when the plugin is installed and a skill is invoked.

What runs on your machine

  • plugins/<name>/skills/<skill>/SKILL.md is markdown read by the agent — no execution.
  • plugins/<name>/skills/<skill>/scripts/ and plugins/<name>/skills/_shared/scripts/ may contain Bash or dependency-free Node.js programs that the agent invokes. Read them before installing a plugin.
  • plugins/memento/ reads and writes files inside the resolved Memento root and may invoke git against that root.
  • plugins/actuary/ is read-only by design; it reads skill files and may use local shell utilities to measure or grep them.

Bundled helper scripts do not make network requests on their own. Skill workflows can direct the host agent to use tools it already has (for example, gh, curl, or a web tool); review the skill instructions and allowed tools before invoking one.

Reporting issues

For non-sensitive issues, open a GitHub issue at https://github.com/richhaase/toolshed/issues. For anything that warrants private disclosure, reach the maintainer via the contact info on https://github.com/richhaase.

There aren't any published security advisories