The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted user-generated content (posts, comments, and livestream interactions) from community platforms via the ae-cli tool. This creates a potential surface for indirect prompt injection where malicious instructions embedded in the analyzed text could influence agent behavior.
Ingestion points: External data enters the agent context through commands like ae-cli community search_posts and get_post_detail in files such as references/community-analyzing-theme-comment.md.
Boundary markers: The instructions include guidelines to wrap typical user feedback in quotes and provide structured sections for "Typical original voice" or "Typical player soundtrack," which serves as a delimiter to help the agent differentiate data from instructions.
Capability inventory: The skill's primary capabilities involve data retrieval and reporting via the ae-cli tool. It lacks tools for executing arbitrary code or performing unauthorized file system writes.
Sanitization: The skill includes explicit filtering rules in references/community-analyzing-theme-comment.md to eliminate "meaningless content," "pure emoticons," and "system messages" before processing, providing a basic layer of data validation.

te-community