The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The documentation in SKILL.md instructs the agent to run sudo apt-get install for various utilities (pandoc, libreoffice, poppler-utils) if they are missing. Encouraging the use of elevated privileges for environment setup is a security risk.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted .docx files, converting them to Markdown or raw XML. This creates a surface where embedded malicious instructions could target the agent.
Ingestion points: DOCX files are converted to Markdown via pandoc or unpacked to raw XML for processing.
Boundary markers: Absent; the instructions do not specify the use of delimiters or 'ignore' warnings for the content extracted from documents.
Capability inventory: The skill uses shell commands for pandoc, python, soffice, and pdftoppm to manipulate files and generate images.
Sanitization: The skill uses defusedxml to mitigate XXE attacks during XML parsing, but no general sanitization of document text is mentioned.
[COMMAND_EXECUTION]: The skill relies on executing external command-line tools like pandoc, libreoffice, and pdftoppm to perform document conversions and image processing.
[EXTERNAL_DOWNLOADS]: Setup instructions include downloading and installing software from official system and package registries (APT and NPM).

docx