The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external scripts and binaries for document verification and rendering. It invokes a bundled Python script (check_office.py) and a Node.js-based LibreOffice CLI tool. The instructions emphasize using absolute paths for these bundled tools rather than searching the system environment or installing new packages, which limits the risk of executing unauthorized binaries.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and manipulate content from external .docx files, which creates a surface for indirect prompt injection if those files contain malicious instructions meant to be interpreted by the agent.
Ingestion points: The skill reads paragraphs, tables, runs, and metadata from user-provided Word documents in the workspace (SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters or "ignore instructions" directives when reading document content.
Capability inventory: The skill has the ability to write to the file system, execute pre-defined Python scripts, and perform document conversion/rendering via shell commands.
Sanitization: There is no evidence of content sanitization or instruction filtering for the text extracted from documents.

office-docx