The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py dynamically generates C source code and executes gcc via subprocess.run to compile it into a shared library at runtime.
[COMMAND_EXECUTION]: The skill utilizes the LD_PRELOAD environment variable to inject a compiled shared library into the soffice process. This process-hooking technique is used to intercept system calls like socket, listen, and accept to bypass environment restrictions.
[COMMAND_EXECUTION]: The script scripts/recalc.py executes soffice with parameters designed to trigger a dynamically configured LibreOffice Basic macro (RecalculateAndSave).
[COMMAND_EXECUTION]: The scripts scripts/office/pack.py and validate.py use RedliningValidator, which executes git diff via subprocess.run to compare document versions.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface through the following evidence chain: 1. Ingestion points: Untrusted data is read from spreadsheets via pd.read_excel and load_workbook (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: The skill can execute system commands and compile code via subprocess.run and gcc (soffice.py). 4. Sanitization: External content is processed without validation or filtering.

xlsx