The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill documentation provides instructions to execute remote code via bunx mcp-remote https://bgpt.pro/mcp/sse. The bunx command functions as a download-and-execute mechanism for Node.js/Bun packages. The target endpoint bgpt.pro and the package bgpt-mcp originate from a third-party source without verifiable provenance or version pinning.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from a scientific paper database. This creates a surface for indirect prompt injection, as the agent retrieves structured full-text content (methods, results, conclusions) from a remote API that could be manipulated to include malicious instructions.
Ingestion points: Data fields returned by the search_papers MCP tool (SKILL.md).
Boundary markers: Absent; the instructions do not implement delimiters or safety warnings for the processed paper data.
Capability inventory: Scientific literature searching and structured data extraction (SKILL.md).
Sanitization: No validation or filtering is specified for the content returned by the remote server.

bgpt-paper-search