The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains commands to download and execute an installation script for the uv package manager from Astral's infrastructure (https://astral.sh/uv/install.sh | sh). This is a common pattern for bootstrapping development tools from well-known sources.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of the tooluniverse Python package from PyPI using the uvx tool. It also provides instructions for cloning the official plugin source from the author's GitHub repository (github.com/mims-harvard/ToolUniverse.git).
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands to manage the environment, including CLI tools for OpenAI Codex (codex), Python dependency management (uv, uvx), and version control (git).
[INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface for indirect prompt injection because it integrates an MCP server that discovers and executes a vast library of tools (2,700+) and research skills (160+). These tools process external data from scientific databases and research platforms.
Ingestion points: Data enters the agent's context through find_tools, get_tool_info, and execute_tool operations performed by the MCP server.
Boundary markers: No explicit boundary markers or instruction-blocking delimiters are specified for tool outputs within the skill instructions.
Capability inventory: The skill enables the agent to perform complex file operations and network requests via the underlying tools provided by the tooluniverse package.
Sanitization: There are no instructions regarding the sanitization or validation of the scientific data or tool outputs before they are processed by the agent.

tooluniverse-codex-plugin