The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts recalc.py and render.py use subprocess.run to invoke system utilities such as soffice (LibreOffice) and pdftoppm (Poppler). These calls are essential for spreadsheet recalculation and rendering sheets to images, which are the primary functions of the skill.
[DYNAMIC_EXECUTION]: The recalc.py script executes a hardcoded Python snippet via subprocess.run([sys.executable, "-c", ...]). This worker script is used to isolate the ironcalc calculation engine, providing process stability and isolation from the main agent environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted .xlsx and .csv files. Mandatory Evidence Chain: 1. Ingestion points: profile.py, audit.py, and recalc.py load Excel and CSV files. 2. Boundary markers: The scripts output structured JSON and text reports to stdout for agent interpretation. 3. Capability inventory: Includes filesystem writes (insert.py) and subprocess execution (recalc.py, render.py). 4. Sanitization: audit.py tokenizes formulas and checks for dangerous functions (e.g., WEBSERVICE, FILTERXML), and insert.py uses hardened XML parsing.
[SAFE]: The insert.py script implements XML parsing using lxml.etree.XMLParser with resolve_entities=False and no_network=True, effectively mitigating XML External Entity (XXE) vulnerabilities during workbook manipulation.

xlsx