The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to use npx to download and run the bgpt-mcp package and the mcp-remote utility from the NPM registry. These are external dependencies not bundled with the skill code.
[REMOTE_CODE_EXECUTION]: The setup process involves running code downloaded via npx, and the primary functionality relies on connecting to a remote MCP server at https://bgpt.pro/mcp/sse, which executes server-side logic to provide tool responses to the agent.
[COMMAND_EXECUTION]: Setup and alternative configuration require the user to execute shell commands involving npx directly in their terminal.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Structured data containing methods, results, and conclusions is ingested from the remote BGPT server via the search_papers tool.
Boundary markers: The skill documentation does not mention any delimiters or instructions for the agent to ignore embedded instructions within the retrieved scientific data.
Capability inventory: The skill environment permits Bash tool usage and requires network access, which could be exploited if malicious content is processed by the agent.
Sanitization: There is no mention of filtering, validation, or sanitization of the content extracted from scientific papers before it is presented to the agent.

alterlab-bgpt-search