The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to use npx to fetch and execute packages (bgpt-mcp, mcp-remote) from the npm registry, which are not provided by a recognized trusted organization.
[REMOTE_CODE_EXECUTION]: Provides instructions to run npx mcp-remote https://bgpt.pro/mcp/sse, which executes code to establish a connection with a remote Server-Sent Events (SSE) endpoint hosted on external infrastructure.
[DATA_EXFILTRATION]: The skill requires network communication with the domain bgpt.pro to search and retrieve paper data, which is an external, non-whitelisted destination.
[COMMAND_EXECUTION]: Includes shell commands (npx bgpt-mcp, npx mcp-remote) for setting up the MCP server environment via the Bash tool.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting structured experimental data from an external source.
Ingestion points: Data returned from the search_papers tool via the remote SSE endpoint https://bgpt.pro/mcp/sse.
Boundary markers: None identified in the provided instructions to delimit or ignore instructions within the retrieved paper content.
Capability inventory: The skill is authorized to use the Bash tool, which could be exploited if injection occurs.
Sanitization: No evidence of sanitization or validation of the remote paper data is provided.

bgpt-paper-search