The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it ingests and processes untrusted data from external .docx files.
Ingestion points: File reading and text extraction via pandoc or raw XML unpacking as described in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands were found in the processing workflow.
Capability inventory: The skill performs subprocess execution (pandoc, soffice, pdftoppm), file system writes, and package management operations.
Sanitization: No sanitization or filtering of extracted document text is implemented before it is processed by the agent.
[COMMAND_EXECUTION]: The skill utilizes several CLI utilities and custom scripts for document processing, including pandoc for text extraction, soffice (LibreOffice) for format conversion, and pdftoppm (Poppler) for rendering. It also executes local Python scripts (accept_changes.py, comment.py) and Node.js scripts to generate and modify document structures.
[EXTERNAL_DOWNLOADS]: Instructions in SKILL.md direct the agent to install the docx library from the NPM registry to facilitate document generation. This is a standard dependency for the skill's stated purpose.

docx-custom