kol-discovery
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from external social media platforms (TikTok, Instagram, YouTube, Xiaohongshu), which may contain hidden instructions targeting the LLM.
- Ingestion points: Data enters the agent context through search and profile tools defined in
SKILL.mdandreferences/sandbase-api-map.md(e.g.,tiktok_app_v3_creator_info,instagram_v3_user_posts). - Boundary markers: The skill lacks explicit instructions or delimiters to help the agent distinguish between its system instructions and the data fetched from external social media sources.
- Capability inventory: The skill utilizes the
sandbase_call_toolcapability to perform network-based data retrieval from external sources. - Sanitization: There is no evidence of sanitization, filtering, or validation of the content retrieved from creator profiles or posts before it is presented to the agent.
Audit Metadata