The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and synthesize data from untrusted external sources, including social media platforms, review sites, and interview transcripts.
Ingestion points: The skill reads external data from Reddit, G2, YouTube, and various other community platforms as described in SKILL.md and references/source-guides.md.
Boundary markers: The instructions do not specify the use of delimiters or safety wrappers when processing external content, increasing the risk that embedded malicious instructions could be followed by the agent.
Capability inventory: The skill is intended to be used by an agent that may have access to tools for file operations, web browsing, and code execution, providing a potential impact path for successful injections.
Sanitization: No sanitization or escaping mechanisms are described for the 'money quotes' and verbatim language the skill is instructed to extract.

customer-research