The Agent Skills Directory

[DYNAMIC_EXECUTION]: The recalc.py script dynamically generates a LibreOffice Basic macro and writes it to the user's application configuration directory (e.g., ~/.config/libreoffice/...). This macro is then executed via soffice to perform formula recalculation. This process can overwrite existing user macros in Module1.xba if they do not contain the expected function string.
[COMMAND_EXECUTION]: The skill uses subprocess.run in recalc.py to execute system-level commands, specifically soffice for spreadsheet processing and timeout/gtimeout for execution control. While these are used for the skill's primary purpose, they represent high-privilege capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external spreadsheet files (.xlsx, .csv, .tsv), creating a surface where malicious instructions embedded in data could influence the agent's behavior.
Ingestion points: recalc.py (via load_workbook on lines 131 and 155) and various Python snippets in SKILL.md using pandas.read_excel and openpyxl.load_workbook.
Boundary markers: None identified; there are no instructions to the agent to ignore or delimit embedded content within the processed files.
Capability inventory: The skill has the ability to write to the file system (wb.save), execute system commands (subprocess.run), and generate/run macros.
Sanitization: No validation or sanitization of spreadsheet cell values is performed before they are used in calculations or written back to files.
[METADATA_POISONING]: There is a discrepancy between the provided author ('kastalien-research') and the ownership statement in LICENSE.txt ('Anthropic, PBC'). This misleading metadata could cause a user to misjudge the skill's origin and safety profile.

xlsx