The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The setup.sh and env_check.sh scripts are designed to download the .NET SDK installation script from https://dot.net/v1/dotnet-install.sh, an official Microsoft domain.
[REMOTE_CODE_EXECUTION]: The tool automates the retrieval and execution of a setup script from a well-known source (dot.net) as part of its environment initialization.
[COMMAND_EXECUTION]: The skill invokes several external utilities to facilitate document operations:
soffice is utilized by doc_to_docx.sh for headless conversion of legacy .doc files.
pandoc is used in docx_preview.sh to extract and display document text for previewing purposes.
dotnet is used to build the core project and execute C# scripts.
[PRIVILEGE_ESCALATION]: The setup.sh script employs sudo to install dependencies through system package managers and applies chmod +x to make the downloaded installation script executable.
[DYNAMIC_EXECUTION]: The skill's documentation provides patterns for writing and executing custom C# scripts (.csx files) at runtime using the dotnet run command to manage advanced structural document manipulations.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an inherent vulnerability surface for indirect prompt injection during its processing of document data:
Ingestion points: Untrusted content enters the agent's context through user-provided .docx and .doc files during analysis, editing, and template application phases.
Boundary markers: The document processing logic does not implement specific delimiters or warnings to separate data content from potential instructions.
Capability inventory: The skill has broad capabilities, including executing shell commands, performing file system operations, and running dynamic scripts.
Sanitization: While the tool performs XML escaping for text insertion, it lacks specific sanitization logic to detect or filter out prompt injection patterns embedded within the input documents.

docx-toolkit