The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation and process injection to maintain functionality in environments with restricted socket access.
File: scripts/office/soffice.py contains logic to write C source code to a temporary file (lo_socket_shim.c), compile it into a shared object (lo_socket_shim.so) using gcc, and then use the LD_PRELOAD environment variable to inject this library into the soffice executable.
[COMMAND_EXECUTION]: Several scripts execute external binaries with varying levels of risk.
File: scripts/office/soffice.py executes gcc for runtime compilation.
File: scripts/accept_changes.py and scripts/office/soffice.py execute soffice (LibreOffice) to process documents.
File: scripts/office/validators/redlining.py executes git diff to compare text content.
File: SKILL.md provides documentation suggesting the use of sudo apt install for system dependencies, which is a best practice violation in automated agent environments.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted .docx files.
Ingestion points: scripts/merge_runs.py, scripts/accept_changes.py, and scripts/office/validate.py all accept external file paths as input.
Boundary markers: Absent. There are no explicit instructions within the skill body to ignore or isolate instructions embedded in the processed document text.
Capability inventory: The skill possesses file write capabilities and the ability to execute shell commands via subprocess.run in soffice.py and accept_changes.py.
Sanitization: The skill uses defusedxml to mitigate standard XML-based attacks like XXE (XML External Entity) injection, which provides a layer of data-level security.

docx