The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes AI-powered features that are susceptible to indirect prompt injection when processing post content.
Ingestion points: The save_post hook and the my_plugin_generate_summary_handler function in SKILL.md process untrusted content from the $post->post_content field.
Boundary markers: Absent. The skill concatenates untrusted content directly into the prompt string (e.g., 'Summarize in 2 sentences: ' . substr($content, 0, 1000)) without using delimiters or instructions to ignore embedded commands.
Capability inventory: The skill possesses capabilities to write to the WordPress database via wp_update_post and perform network-based AI requests using the wp_ai_client_prompt function.
Sanitization: The implementation uses strip_tags() for input and sanitize_textarea_field() for output. While these prevent XSS and HTML injection, they do not mitigate the risk of an LLM following malicious instructions embedded within the plain text content.

wordpress