The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform site builds and previews using bundle exec jekyll build and bundle exec jekyll serve. These commands execute site-specific Ruby plugins and configuration. The skill mitigates this risk by providing explicit security warnings, stating that "a build is not a sandbox for untrusted project code," and requiring the agent to verify execution boundaries and destination directories before execution.
[EXTERNAL_DOWNLOADS]: The skill references https://rubygems.org for managing Ruby gem dependencies like Jekyll. It guides the agent to use established project environments (e.g., Gemfile.lock) and warns against arbitrary tool installation or global configuration changes. Downloads from official package registries for project-specific dependencies are standard developer workflows.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of Markdown content, Liquid templates, and data files. It provides guidance on using native Liquid filters such as escape, jsonify, and xml_escape to sanitize output and includes warnings that these filters do not validate arbitrary scripts. This ingestion surface is intrinsic to the skill's purpose as a site generator.

jekyll