The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py contains embedded C source code that is written to a temporary file and compiled at runtime using gcc. The resulting shared object is then loaded into the environment via the LD_PRELOAD environment variable. While this is used to provide a compatibility layer for socket operations in restricted environments, it involves dynamic code generation and process injection techniques that are high-risk.
[COMMAND_EXECUTION]: Multiple scripts, including accept_changes.py, pack.py, unpack.py, and validate.py, use the subprocess module to execute external binaries such as soffice, pandoc, git, and gcc. This broad capability surface allows the skill to interact deeply with the host system.
[PROMPT_INJECTION]: The skill processes untrusted document content from .docx files through unpack.py and pandoc. There are no explicit boundary markers or instructions provided to the agent to treat the ingested document text as untrusted data, which creates a surface for indirect prompt injection attacks where a malicious document could influence the agent's behavior.
[EXTERNAL_DOWNLOADS]: The SKILL.md file instructs the agent to install external dependencies including the docx Node.js library and system tools like pandoc, LibreOffice, and Poppler utilities. These are necessary for the skill's primary functionality but involve downloading and installing third-party software.

docx