The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a workflow where the agent writes C# code to Program.cs at runtime, which is then compiled and executed using the .NET SDK via the ./scripts/docx build command. This dynamic code generation and execution path can be risky if untrusted input influences the generated code.
[COMMAND_EXECUTION]: The skill includes and executes a pre-compiled ELF binary located at validator/Validator. This binary is used to validate OpenXML structures. Shipping and running opaque binaries is a security concern as their internal logic is difficult to audit.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses user-uploaded .docx files by unzipping them and reading the internal XML content. This creates a surface for indirect prompt injection, where malicious instructions inside a document could potentially influence the agent's behavior during the editing or code-generation process.
Ingestion points: scripts/docx_lib/editing/context.py (unzips and parses user-provided docx files).
Boundary markers: Absent; the skill does not use specific delimiters or warnings to ignore instructions embedded in the document XML.
Capability inventory: Compilation and execution of generated C# code via dotnet, execution of the validator/Validator binary, and extensive file system write operations.
Sanitization: The skill uses xml_escape when interpolating text into XML structures in scripts/docx_lib/editing/revisions.py, providing protection against XML injection but not necessarily against natural language prompt injection.
[EXTERNAL_DOWNLOADS]: The skill's initialization script (scripts/docx) downloads the .NET SDK installation script from Microsoft's official dot.net domain. This is an expected operation for environment setup and targets a well-known, trusted service provider.

kimi-docx