The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs runtime compilation and process injection in scripts/office/soffice.py.
It contains hardcoded C source code for a socket shim.
At runtime, it writes this source to a temporary file and executes gcc to compile a shared object (.so).
It uses the LD_PRELOAD environment variable to inject this compiled library into the LibreOffice (soffice) process to bypass socket restrictions.
[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to execute system commands.
It invokes soffice for document conversion and change acceptance in scripts/accept_changes.py and scripts/office/soffice.py.
It executes gcc for the runtime compilation mentioned above.
SKILL.md instructs the agent to run pandoc and pdftoppm via the command line.
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md explicitly instructs the user to install an external package globally using npm install -g docx.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection (Category 8).
Ingestion points: The skill reads untrusted content from .docx files using pandoc or by unpacking raw XML.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to distinguish between its own instructions and the content extracted from processed documents.
Capability inventory: The skill possesses powerful capabilities including arbitrary file writing (scripts/office/unpack.py), network-related process shimming (scripts/office/soffice.py), and complex subprocess execution.
Sanitization: While it uses defusedxml for some XML operations, the overall workflow involves the agent reading and potentially acting upon the natural language content of external documents.

docx