The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands at runtime using the subprocess.run module.
scripts/office/soffice.py invokes gcc to compile a C shared library intended for system call interception.
scripts/recalc.py and scripts/office/soffice.py execute the soffice (LibreOffice) binary.
scripts/office/validators/redlining.py executes git diff for document comparison.
scripts/recalc.py performs a persistent configuration change by writing a StarBasic macro to the user's LibreOffice application directory (e.g., ~/.config/libreoffice/4/user/basic/Standard/Module1.xba), which runs automatically when triggered by the skill's recalculation logic.
[REMOTE_CODE_EXECUTION]: The skill utilizes high-risk dynamic execution techniques to run and inject code at runtime.
scripts/office/soffice.py contains hardcoded C source code (_SHIM_SOURCE) that is written to a file, compiled at runtime using gcc, and loaded into the soffice process using the LD_PRELOAD environment variable. This allows the interception and modification of low-level system calls (socket, listen, accept, close), facilitating code injection and the potential bypass of security sandbox restrictions.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface (Category 8).
Ingestion points: It is designed to read and process external spreadsheet files (.xlsx, .csv, .tsv, .xlsm) using pandas and openpyxl (specified in SKILL.md), which may contain malicious instructions in cells or metadata.
Boundary markers: The skill does not implement delimiters or provide specific instructions to the agent to ignore potentially adversarial commands embedded in the processed data.
Capability inventory: The skill possesses powerful system-level capabilities across multiple scripts, including shell command execution (subprocess.run in recalc.py, soffice.py, redlining.py) and the ability to compile and inject binaries.
Sanitization: Content from processed spreadsheets is not sanitized or validated before being incorporated into the agent's context or workflow, creating a risk that malicious spreadsheet data could influence agent actions.

xlsx