The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates downloading a Python binary wheel (tuplatform_connect-0.3.0-py3-none-any.whl) and installation requirements (install-requirements.txt) from connect.aiscientist.tools. This domain is not identified as a trusted organization or well-known service in the security configuration.
[REMOTE_CODE_EXECUTION]: External components downloaded from the aforementioned domain are installed and executed via pip. While the wheel installation includes a SHA-256 checksum for integrity verification, the source remains an unverified third-party domain, posing a risk of executing malicious remote code.
[INDIRECT_PROMPT_INJECTION]: The run_compass_prediction tool ingests data from external files specified by the user through the gene_expression_data_path parameter, creating a potential vulnerability surface.
Ingestion points: The tool reads content from gene expression data files (e.g., TSV format) provided at runtime.
Boundary markers: The skill documentation does not specify the use of delimiters or 'ignore' instructions to separate data from instructions within the processed files.
Capability inventory: The skill facilitates setting up a Model Context Protocol (MCP) server that executes Python code and shares tools through an external relay service.
Sanitization: There is no evidence of input validation or sanitization mechanisms to filter or escape content from the ingested data files.
[DYNAMIC_EXECUTION]: The skill provides example code blocks that use shell heredocs to dynamically generate and execute Python scripts to verify MCP tool discovery and execution.

setup-immune-compass-remote-tool