The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill performs runtime code generation and process injection via a pre-loading mechanism.
Evidence: The script scripts/office/soffice.py writes a C source code string to a temporary file, compiles it into a shared library using gcc, and then sets the LD_PRELOAD environment variable to load this library when launching LibreOffice. This is used as a shim to redirect AF_UNIX socket calls in restricted environments.
[COMMAND_EXECUTION]: The skill executes multiple system commands using Python's subprocess module.
Evidence: Multiple scripts (scripts/recalc.py, scripts/office/soffice.py, scripts/office/validators/redlining.py) call system utilities including gcc, git, and soffice with various arguments.
[PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection from untrusted spreadsheet data.
Ingestion points: Data from spreadsheet files is loaded into the agent context via pandas and openpyxl as instructed in SKILL.md.
Boundary markers: Absent; there are no delimiters or explicit instructions to ignore embedded commands in the processed tabular data.
Capability inventory: The skill has the ability to execute shell commands and perform file system writes.
Sanitization: Absent; external data is processed and potentially used in further automated reasoning or file operations without filtering.

xlsx