The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents and enables the execution of arbitrary PHP code on the server through the php_exec attribute.
Evidence: The 'How It Renders' section describes PHP execution via temporary file inclusion when php_exec is enabled.
Gating: Execution is restricted by a site-level constant (MD_ALLOW_PHP_SNIPPETS) and a content checksum verification, which provides some protection but does not prevent a malicious agent from intentionally pushing malicious PHP code.
[INDIRECT_PROMPT_INJECTION]: The skill manages blocks that ingest and render arbitrary HTML and JavaScript, creating a potential surface for cross-site scripting (XSS) or prompt injection if the source data is compromised.
Ingestion points: The content, css, and js fields in the POST /wp-json/md/v1/page-blocks endpoint and the marketers-delight/inline-page-block Gutenberg block attributes.
Capability inventory: Server-side PHP execution via php_exec and client-side JavaScript execution through the js field.
Boundary markers: The documentation does not specify boundary markers to prevent the agent from obeying instructions embedded within the HTML/JS content it processes.
Sanitization: While a checksum is used to verify PHP integrity against database tampering, there is no evidence of semantic sanitization or filtering for the content, CSS, or JavaScript being pushed via the API.

page-block