The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to execute packages from the npm registry using npx (bgpt-mcp and mcp-remote). These are third-party components that are downloaded at runtime.
[REMOTE_CODE_EXECUTION]: The use of npx for MCP configuration involves downloading and running code from external repositories. The security of this operation depends entirely on the integrity of the third-party package maintainer.
[DATA_EXFILTRATION]: The skill requires active internet access to bgpt.pro and transmits user search queries to this external domain to retrieve results.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process full-text scientific papers and structured experimental data from external sources, creating a surface for indirect injection.
Ingestion points: Text and data fields returned by the search_papers MCP tool.
Boundary markers: The instructions lack explicit delimiters or system-level directives to ensure the agent ignores any potential instructional content embedded within the retrieved scientific papers.
Capability inventory: No local scripts or executables are included in the skill; execution is limited to the configured MCP tools.
Sanitization: There is no mention of sanitization, filtering, or validation of the content retrieved from the external service before it is presented to the agent context.

bgpt-paper-search