The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to systematically collect and analyze a large dataset (500-1000+) of social media posts from external creators to identify viral patterns (references/reverse-engineering.md).
Ingestion points: Untrusted data enters the agent's context through the scraping or collection of external social media content as described in the 'Scrape' step of the framework.
Boundary markers: The instructions lack specific delimiters or warnings to ignore potentially malicious instructions embedded within the ingested social media content.
Capability inventory: While the skill itself does not contain scripts, it is designed for an agent that may possess tools for web browsing, file system access, or network communication, which could be exploited if an ingested post contains a successful injection.
Sanitization: There are no instructions for sanitizing or filtering the content fetched from external platforms before the agent processes it for analysis.

social-content