The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs network operations to interact with an external service at mcp.tikhub.io which is not on the default whitelist.
Evidence: scripts/tikhub_mcp.py uses urllib.request to communicate with the TikHub MCP API. It sends the TIKHUB_API_KEY (sourced from environment variables) as a Bearer token in the request headers to https://mcp.tikhub.io/.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as its primary function involves processing untrusted user-generated content from various social media platforms.
Ingestion points: The skill is designed to fetch and analyze posts, comments, and transcripts from platforms like TikTok, Douyin, X, and Reddit via the tikhub_mcp.py script.
Boundary markers: While the skill includes high-level safety guidelines in references/safety.md, it lacks explicit technical delimiters or validation steps to isolate instructions embedded within the social media data it processes.
Capability inventory: The skill has the ability to make network requests and perform local file system writes, which could be leveraged if the agent inadvertently follows instructions found in processed social media content.
Sanitization: Data normalization scripts (scripts/normalize.py) focus on structural mapping and do not include content sanitization or filtering logic.

pongfi-research