The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of social media post content, which could be used as a vector for injection attacks.
Ingestion points: The post_content input field in SKILL.md accepts external text.
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to treat the post_content as data rather than instructions.
Capability inventory: The skill uses refly skill run and refly workflow CLI commands to process the data and interact with the Refly platform.
Sanitization: There is no evidence of input validation or sanitization for the content before it is passed to the workflow execution.

facebook