The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data fetched from external social media platforms (Reddit, Twitter, Instagram, Xiaohongshu), which are attacker-controlled sources that may contain malicious instructions designed to influence the agent.
Ingestion points: Social media content is ingested via tools like reddit_app_dynamic_search and twitter_web_search_timeline (SKILL.md).
Boundary markers: The skill includes instructions to cite sources and separate factual observations from analysis, which helps contextualize external data, though it lacks explicit structural delimiters for tool outputs.
Capability inventory: The skill utilizes sandbase_call_tool to interact with search APIs; it does not possess file-write or system-level execution capabilities.
Sanitization: There is no explicit evidence of content filtering or sanitization for the retrieved social media data.

audience-research