The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py performs runtime compilation and process injection. It writes an embedded C source string to a temporary file, compiles it using gcc, and then loads the resulting shared library into the soffice process via the LD_PRELOAD environment variable. This technique is used to shim system calls but represents a high-risk dynamic execution pattern.
[COMMAND_EXECUTION]: Multiple scripts within the skill utilize subprocess.run to execute external system utilities. This includes gcc for compiling the shim, soffice for spreadsheet recalculation, and git for document diffing and validation.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests and processes data from untrusted Excel and CSV files. These files may contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Data is imported via pandas.read_excel and openpyxl.load_workbook as described in SKILL.md and used in scripts/recalc.py.
Boundary markers: The skill does not implement explicit markers or instructions to isolate untrusted spreadsheet data from the agent's operational logic.
Capability inventory: The skill has extensive capabilities, including file system access and the ability to execute shell commands via subprocess.run.
Sanitization: The skill employs defusedxml for XML parsing in some components, which helps mitigate certain XML-based attacks.

xlsx