The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external influencer profiles on platforms like TikTok and Instagram, which constitutes an indirect prompt injection surface. However, the risk is assessed as safe given the primary purpose is research and the skill lacks dangerous capabilities.
Ingestion points: Influencer performance metrics, channel descriptions, and profile data from external social platforms (SKILL.md).
Boundary markers: The skill includes instructions to "Cite sources and preserve attribution" and to "Separate factual observations from analysis," which encourages the agent to maintain context separation (SKILL.md).
Capability inventory: The skill is limited to descriptive research and data synthesis using the SandBase gateway; it does not request or use capabilities for file writing, shell command execution, or arbitrary network access.
Sanitization: No explicit sanitization or filtering of external profile data is mentioned.
[SAFE]: The skill uses vendor-specific tools (sandbase_describe_tool, sandbase_call_tool) that match the skill's stated purpose and author (sandbaseai). The workflow follows a defensive "describe-then-call" pattern to ensure schema compliance.

influencer-analytics