The Agent Skills Directory

[PERSISTENCE]: The skill instructs both users and agents to modify shell profile files, such as ~/.zshrc or ~/.bashrc, to persist API keys as environment variables across sessions (found in SKILL.md and TROUBLESHOOTING.md).
[PRIVILEGE_ESCALATION]: Troubleshooting documentation (TROUBLESHOOTING.md) suggests using sudo to create symbolic links in protected system directories like /usr/local/bin or /opt/homebrew/bin to resolve path issues for the uv tool.
[DYNAMIC_EXECUTION]: The diagnostic and verification scripts (scripts/diagnose_setup.py and scripts/verify_installation.py) employ the __import__ function to dynamically load and check for the presence of optional library dependencies such as rdkit and biopython at runtime.
[COMMAND_EXECUTION]: Multiple utility scripts use subprocess.run and subprocess.Popen to execute system commands for environmental discovery, including checking versions of uv, pip, and python, as well as validating that the ToolUniverse CLI tools are functional.
[EXTERNAL_DOWNLOADS]: The skill fetches the uv package manager installer from astral.sh (a well-known service) and downloads research workflows/skills from the official mims-harvard/ToolUniverse GitHub repository. These actions are documented neutrally as they originate from trusted or well-known entities.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect injection surface by instructing the agent to read external instructions from https://aiscientist.tools/setup.md and by directing the agent to write its own configuration and memory rules into platform-specific files like CLAUDE.md, .cursor/rules/tooluniverse.md, and .clinerules/tooluniverse.md.

setup-tooluniverse