The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches installation requirements and configuration files from https://connect.aiscientist.tools/install-requirements.txt and the mims-harvard GitHub repository. These files are used to define the environment and dependencies for the tool.
[REMOTE_CODE_EXECUTION]: The setup instructions include installing a Python wheel directly from a remote URL: https://connect.aiscientist.tools/downloads/tuplatform_connect-0.3.0-py3-none-any.whl. Although the skill includes a SHA-256 hash for integrity verification, installing executable code from non-standard package registries increases the risk of supply chain attacks.
[COMMAND_EXECUTION]: The skill requires extensive use of the shell to create virtual environments, install packages via uv and pip, and manage background services using a custom CLI tool (tu). It also executes multiple Python scripts to verify the functionality of the local MCP (Model Context Protocol) server.
[INDIRECT_PROMPT_INJECTION]: The tool is designed to process biological sequences provided as external input (e.g., {"sequence":"MKT"}). This creates an attack surface where maliciously crafted sequence data could potentially exploit vulnerabilities in the underlying ESM model or the tool's processing logic. The skill notes the need for OOM and timeout sanitization, acknowledging these risks.
[CREDENTIALS_UNSAFE]: The instructions reference the management of sensitive authentication data, including TOOLUNIVERSE_SERVICE_KEY and TOOLUNIVERSE_API_TOKEN. While it advises using environment variables or protected files rather than command-line arguments, it directs the agent to interact with sensitive files like tooluniverse-service.env.

setup-esm-remote-tool