The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a research process that relies on ingesting data from untrusted external environments, creating a susceptibility to indirect prompt injection.
Ingestion points: The SKILL.md file directs the agent to gather information from potentially high-risk sources, including social media, wikis, and community discussion forums.
Boundary markers: There are no instructions for the agent to use delimiters or to treat the retrieved content as data rather than instructions, which would help prevent the agent from obeying malicious commands embedded in the search results.
Capability inventory: The skill's primary function requires the agent to utilize search or web browsing tools to access these external sources.
Sanitization: The skill does not mention any requirements for filtering, validating, or sanitizing the retrieved data before it is presented to the user or incorporated into the final report.

creative-research