The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires elevated system privileges on Linux via sudo apt-get for software installation. It also relies on extensive shell command execution for environment detection, package management, and application control.
[REMOTE_CODE_EXECUTION]: Fetches and executes a shell script directly from the well-known Ollama service via a shell pipe. It also installs various third-party software from GitHub repositories and public registries (PyPI, NPM).
[COMMAND_EXECUTION]: Employs dynamic code execution by generating custom Python harness scripts at runtime using write_file and installing them as local editable packages via pip install -e.
[EXTERNAL_DOWNLOADS]: Automatically installs a wide array of desktop applications (GIMP, LibreOffice, Blender, etc.) using system package managers like winget, brew, and apt without prior verification of the package content.
[PROMPT_INJECTION]: The skill creates a significant indirect prompt injection surface by processing untrusted data (such as documents, spreadsheets, and 3D scenes) and executing internal scripts or macros found within them (e.g., Blender scripts, Excel macros, Jupyter notebooks). This ingestion of untrusted data into executable contexts lacks boundary markers or sanitization logic.

cli-anything