The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by design. It facilitates reading existing content from a WordPress database and provides the agent with powerful capabilities to modify site-wide settings and content based on that data.
Ingestion points: The tools woodmart/gutenberg-get-content and woodmart/search-content allow the agent to pull arbitrary text from the site into its context window.
Boundary markers: While the skill provides clear instructions on how to handle content, it lacks technical delimiters (like XML tags or specific delimiters) to separate processed data from internal instructions.
Capability inventory: The skill allows creating, renaming, and publishing content (woodmart/create-content, woodmart/update-content), editing site layouts (woodmart/gutenberg-edit-blocks), and changing site-wide theme settings (woodmart/update-theme-setting).
Sanitization: The instructions do not specify input sanitization, relying instead on a manual 'AI Block Queue' admin page for finalization of block-based changes, which acts as a mitigating control.
[SAFE]: The skill enforces a human-in-the-loop safety mechanism through a mandatory finalization process ('Workflow E'). Changes to blocks are queued and require a user to manually approve them in the WordPress admin interface before they go live.
[SAFE]: The instructions explicitly require the agent to seek user confirmation before performing sensitive operations, such as changing the status of a post to 'publish' or modifying slugs on live pages that could result in 404 errors.

woodmart-mcp