The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's setup instructions direct users to execute remote packages using npx mcp-remote and npx bgpt-mcp. This allows for the runtime download and execution of arbitrary code from the npm registry and a remote endpoint.
[EXTERNAL_DOWNLOADS]: The skill relies on external resources hosted at https://bgpt.pro/mcp/sse and a personal GitHub repository connerlambden/bgpt-mcp. These sources are not recognized as trusted organizations or well-known services, increasing the risk of malicious payload delivery.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external data from scientific papers.
Ingestion points: Data is retrieved from full-text studies via the search_papers tool (SKILL.md).
Boundary markers: The instructions lack explicit delimiters or warnings to ignore potentially malicious instructions embedded within the paper data.
Capability inventory: The skill specifies Bash in its allowed-tools configuration, providing a sensitive execution environment that could be targeted by instructions within ingested data.
Sanitization: There is no evidence of sanitization or strict schema validation for the data returned from the remote paper database.

bgpt-paper-search