The Agent Skills Directory

Prompt Injection (HIGH): The tool aptos_intent accepts natural language input to perform blockchain actions. This is a high-risk surface for indirect prompt injection where malicious instructions embedded in processed data could lead to unauthorized asset transfers or swaps. * Ingestion points: Input to mcp__world_u_aptos__aptos_intent tool. * Boundary markers: None specified. * Capability inventory: aptos_transfer, aptos_swap, aptos_stake (high-privilege financial operations). * Sanitization: No sanitization mechanisms are described for natural language processing.
Command Execution (MEDIUM): The skill grants access to tools that modify blockchain state and manage assets. While intended, these are high-privilege capabilities that require strict oversight.
Metadata Poisoning (LOW): The description 'Unicode/typesetting tools' contradicts the actual blockchain-focused toolset, indicating misleading or incorrect metadata.

u