The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze untrusted content from external sources (such as Reddit threads, YouTube transcripts, and listicles) to perform prompt-audits and strategy development. This creates a surface for indirect prompt injection where malicious instructions embedded in those sources could attempt to override the agent's guidelines. * Ingestion points: The skill instructs the agent to 'Read the cited sources' from the web (referenced in references/audit-and-measurement.md). * Boundary markers: The skill does not provide specific instructions to use delimiters or ignore potential commands embedded in the analyzed data. * Capability inventory: The skill orchestrates with other tools like scheduling-and-queue, which handles social media publishing, creating a potential path for external data to influence automated actions. * Sanitization: There are no mentions of filtering or sanitizing the retrieved web content before analysis.

ai-search-optimization