The Agent Skills Directory

[COMMAND_EXECUTION]: Script scripts/office/soffice.py performs runtime compilation of an embedded C source using gcc and utilizes the LD_PRELOAD environment variable to inject the shared object into the LibreOffice (soffice) process. This is used to shim Unix domain sockets. Additionally, scripts such as scripts/recalc.py and scripts/office/validators/redlining.py execute system binaries including soffice and git via subprocess.run with parameters derived from file paths.
[PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection. Ingestion points: Processes untrusted data from user-supplied spreadsheet files (.xlsx, .csv) via openpyxl and pandas. Boundary markers: Absent; spreadsheet cell contents are read into the agent context without isolation markers or instructions to disregard embedded commands. Capability inventory: The skill can execute arbitrary system commands (subprocess.run), write to the file system, and perform runtime code compilation. Sanitization: Absent; no escaping or validation of cell data is implemented before the information is used in the reasoning pipeline.

xlsx