The Agent Skills Directory

[COMMAND_EXECUTION]: The skill spawns several system processes to perform its tasks:
scripts/office/soffice.py executes gcc to compile a shared object shim library.
scripts/recalc.py and scripts/office/soffice.py execute soffice (LibreOffice) to update Excel formulas.
scripts/office/validators/redlining.py executes git diff to perform document version comparisons.
[REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation and environment-based injection:
scripts/office/soffice.py writes C source code (_SHIM_SOURCE) to a temporary file, compiles it at runtime into a shared library, and injects it into the soffice process environment using the LD_PRELOAD mechanism to bypass socket restrictions.
[PROMPT_INJECTION]: The skill processes external data and has an attack surface for indirect prompt injection:
Ingestion points: The skill reads and processes user-provided spreadsheet files (.xlsx, .csv, etc.) as documented in SKILL.md.
Boundary markers: There are no explicit delimiters or safety instructions present to prevent the agent from accidentally obeying instructions embedded within ingested spreadsheet data.
Capability inventory: The skill can perform arbitrary file system writes and execute subprocesses (documented in pack.py and recalc.py).
Sanitization: The skill uses the defusedxml package for XML parsing to mitigate XML-based attacks (XXE), but it does not implement sanitization for natural language instructions in the cell content.

xlsx