The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py dynamically generates C source code, compiles it into a shared library using gcc, and injects it into the LibreOffice process using the LD_PRELOAD environment variable. This is implemented to facilitate communication when AF_UNIX sockets are blocked by a sandbox environment, but the pattern of runtime compilation and process injection is inherently sensitive.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external .pptx files via markitdown and custom XML processing scripts. This creates a vulnerability surface where a malicious presentation could contain hidden instructions designed to override agent behavior during analysis, particularly as the skill has significant capabilities including file system access and subprocess execution.

pptx