The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and synthesize untrusted data from external sources, creating a surface for indirect prompt injection attacks. * Ingestion points: In SKILL.md, the agent is directed to process customer interview transcripts, survey results, support tickets, and content from public platforms detailed in references/source-guides.md such as Reddit, G2, Hacker News, and LinkedIn. * Boundary markers: The instructions lack explicit guidance on using delimiters or 'ignore' instructions when processing external content to prevent the agent from following malicious commands embedded in the data. * Capability inventory: Across SKILL.md and its references, the skill assumes the agent has the capability to read files and access the network to mine data from the specified online platforms. * Sanitization: There are no instructions for sanitizing or validating the ingested content to filter out potentially harmful prompt injection payloads.

customer-research