The Agent Skills Directory

[COMMAND_EXECUTION]: Privilege escalation instructions are present in SKILL.md.
Evidence: The 'Prerequisites' section instructs the agent to execute sudo apt install -y libreoffice if the binary is not found on the system.
[REMOTE_CODE_EXECUTION]: Runtime compilation of C code in scripts/office/soffice.py.
Evidence: The script writes a hardcoded C source string (_SHIM_SOURCE) to a file in the temporary directory and invokes gcc to compile it into a shared object file.
[REMOTE_CODE_EXECUTION]: Library injection using LD_PRELOAD in scripts/office/soffice.py.
Evidence: The script modifies the environment of the soffice process by setting LD_PRELOAD to the path of the dynamically compiled shared library.
[REMOTE_CODE_EXECUTION]: Dynamic macro generation in scripts/recalc.py.
Evidence: The script generates and installs a LibreOffice Basic macro (RECALCULATE_MACRO) into a temporary user profile to automate formula recalculation.
[PROMPT_INJECTION]: Susceptibility to indirect prompt injection from processed Excel files.
Ingestion points: Spreadsheet data is ingested via openpyxl, pandas, and markitdown as specified in SKILL.md and used within scripts/recalc.py.
Boundary markers: Absent; there are no instructions or delimiters to isolate untrusted data from the agent's instructions.
Capability inventory: The skill has significant capabilities including shell command execution and runtime code compilation (scripts/office/soffice.py and scripts/recalc.py).
Sanitization: Absent; the content read from spreadsheets is not validated or sanitized before processing.

xlsx