The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/setup.sh script executes package management commands using sudo to install system-level dependencies. This behavior grants the script administrative privileges on the host system. Evidence: Usage of sudo apt-get, sudo dnf, sudo pacman, and sudo zypper for package installation in scripts/setup.sh.
[EXTERNAL_DOWNLOADS]: The skill downloads an installation script from an external Microsoft domain to set up the .NET environment. Evidence: Fetches https://dot.net/v1/dotnet-install.sh via wget and curl in scripts/setup.sh and scripts/env_check.sh.
[EXTERNAL_DOWNLOADS]: The project initialization process involves fetching external code libraries from a public registry. Evidence: Execution of dotnet restore which connects to nuget.org in scripts/setup.sh and scripts/setup.ps1.
[REMOTE_CODE_EXECUTION]: The skill's primary doc-generation pipeline involves the agent dynamically generating C# code based on user requirements and executing it using the .NET runtime. Evidence: Instructions in SKILL.md to write logic to scripts/dotnet/task.csx and execute via dotnet run.
[REMOTE_CODE_EXECUTION]: A vulnerability surface exists where untrusted user input describing document content could influence the generated C# code, which has broad system capabilities. Ingestion points: User requests for document content in the agent prompt (SKILL.md). Boundary markers: Absent; user data is not clearly delimited from instructions. Capability inventory: Process execution (dotnet run), file system access (File.Copy), and XML manipulation (XDocument.Load) in the MiniMaxAIDocx project. Sanitization: Absent; the agent is not instructed to sanitize or escape user-provided data before embedding it in executable code snippets.

minimax-docx