The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py performs runtime code generation and process injection. It writes a C source file to a temporary directory, compiles it into a shared object using gcc, and then executes LibreOffice (soffice) with the LD_PRELOAD environment variable set to the path of the newly created library. This is used to shim socket-related system calls to work around sandbox restrictions.
[COMMAND_EXECUTION]: The skill frequently executes external system binaries using subprocess.run. Key operations include document conversion via soffice, image generation from PDFs via pdftoppm, and difference analysis via git in scripts/office/validators/redlining.py.
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection (Category 8). It ingests untrusted data from external .pptx and .docx files (identified in SKILL.md and scripts/office/unpack.py). The extracted text is interpolated into the agent's context without the use of boundary markers or specific sanitization routines to isolate embedded instructions from legitimate content.

pptx