The Agent Skills Directory

[SAFE]: The skill defines a specialized safety protocol called 'Read-only mode' designed to prevent unauthorized modifications to the user's environment.
[SAFE]: It mandates the use of a sandboxed scratch directory (mktemp -d) for any necessary write operations, ensuring that temporary files, dependency installs, or code copies do not affect the main workspace without permission.
[SAFE]: The skill explicitly defines what constitutes a 'write' (file edits, Git state changes, package installations, and external side effects like non-GET HTTP requests) and requires an explicit user grant ('unlock', 'apply it') before performing such actions.
[SAFE]: The use of the disable-model-invocation: true frontmatter field is a platform-level restriction that limits the skill's capabilities to static instructions, further reducing the attack surface.

read-only