The Agent Skills Directory

[Command Execution] (LOW): The script executes system commands (soffice, timeout, gtimeout) via subprocess.run to perform its primary function. While it uses list arguments to mitigate shell injection, the reliance on external binaries is a functional necessity.
[Dynamic Execution] (MEDIUM -> LOW): The script dynamically generates a LibreOffice Basic macro (Module1.xba) and writes it to the user's application configuration directory to enable formula recalculation. This behavior is considered LOW severity because it is performed by a trusted author (Anthropic) for a specific, transparent purpose.
[Indirect Prompt Injection] (LOW): The skill possesses a data ingestion surface that processes untrusted Excel files.
Ingestion points: recalc.py reads file content using openpyxl.load_workbook.
Boundary markers: Absent; the script reads cell values directly without delimiters.
Capability inventory: File system access, script generation (macro creation), and execution of local binaries via subprocess.run.
Sanitization: The script scans cell contents for static Excel error strings (e.g., #VALUE!) but does not sanitize or validate general cell content, which could lead to injection if results are later processed by an LLM.
[Persistence Mechanisms] (LOW): The skill installs a persistent macro into the local LibreOffice profile. This modification remains on the system across sessions, which is required for the tool's operation.

xlsx