researchers-biographical
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it is designed to ingest large amounts of untrusted data from the open web using
WebFetchandWebSearch. - Ingestion points: Data is retrieved from interviews, profiles, social media (Twitter, LinkedIn), and news archives as described in
SKILL.mdandINTERVIEW_SOURCES.md. - Boundary markers: The instructions do not include boundary markers or explicit directives for the agent to ignore commands or instructions embedded within the research material.
- Capability inventory: The skill has
Write,Edit,WebFetch, andWebSearchpermissions, allowing it to modify files or perform further network operations based on potentially malicious data. - Sanitization: There is no mention of sanitizing or escaping the fetched content before it is processed by the agent to build the biographical profiles.
- [EXTERNAL_DOWNLOADS]: The skill instructions specifically direct the agent to fetch data from various external platforms and archives, including YouTube, Listen Notes, and social media sites. While these are used for the primary research purpose, they involve retrieving content from untrusted third-party infrastructure.
Audit Metadata