The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py dynamically generates C source code, writes it to a temporary file, and compiles it using gcc into a shared library. This library is then loaded into the soffice process environment via the LD_PRELOAD environment variable to hook system calls for socket management. This runtime compilation and injection of code is a significant security concern.
[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to execute external binaries. In scripts/office/soffice.py, it executes gcc and soffice. In scripts/recalc.py, it executes timeout or gtimeout. In scripts/office/validators/redlining.py, it executes git diff for comparing document versions.
[PROMPT_INJECTION]: The skill ingests and processes untrusted spreadsheet data which creates a vulnerability surface for indirect prompt injection.
Ingestion points: scripts/recalc.py and scripts/office/validate.py load user-provided files via openpyxl and zipfile modules.
Boundary markers: Absent; there are no clear delimiters or instructions to the agent to disregard content found within the spreadsheet data.
Capability inventory: The skill possesses powerful capabilities including arbitrary shell command execution and runtime C compilation.
Sanitization: While defusedxml is used for XML parsing, the skill lacks sanitization for spreadsheet cell content that could be used to influence agent behavior.

xlsx