The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py dynamically generates C code, compiles it using gcc, and uses LD_PRELOAD to inject it into the soffice process to hook system calls. This process injection is a highly sophisticated and high-risk technique. Additionally, scripts/recalc.py writes a StarBasic macro to the local LibreOffice configuration to automate recalculations.
[METADATA_POISONING]: The description in SKILL.md is a misleading error-like string ('QUERY LENGTH LIMIT EXCEEDED'), which obscures the skill's actual purpose and behavior.
[COMMAND_EXECUTION]: The scripts scripts/recalc.py, scripts/office/soffice.py, and scripts/office/validators/redlining.py execute shell commands using subprocess.run (e.g., gcc, soffice, git).
[INDIRECT_PROMPT_INJECTION]: The skill lacks boundary markers and sanitization for data ingested via pd.read_excel (instructions) and load_workbook (scripts/recalc.py). Ingestion points: SKILL.md and scripts/recalc.py. Boundary markers: Absent. Capability inventory: Shell execution and process injection. Sanitization: Absent for natural language data. This combination presents a significant risk from malicious data files.

xlsx