The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Excel workbooks, creating a surface for indirect prompt injection.
Ingestion points: Processes .xlsx and .xlsm files from the {task}/ directory (SKILL.md, Step 1).
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions when reading cell values or formulas.
Capability inventory: The skill executes local Python scripts recalc.py and audit.py (SKILL.md, Steps 2 and 3) and writes report files to the {task}/ directory.
Sanitization: The instructions do not mention sanitizing or validating workbook data before processing it through scripts or including it in the final report.

check-model