The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/office/soffice.py script contains an embedded C source that is compiled at runtime using gcc. The resulting shared library is loaded via LD_PRELOAD to intercept system socket calls when running LibreOffice. While used for environment compatibility, runtime compilation and process injection are high-risk capabilities.
[METADATA_POISONING]: There is an inconsistency between the declared author ('new-career-begin') and the copyright holder specified in the LICENSE.txt file. This mismatch in attribution metadata is potentially misleading.
[INDIRECT_PROMPT_INJECTION]: The skill reads and converts untrusted PowerPoint files, feeding the resulting text and visual data back into the agent context. Without explicit sanitization or instructions to ignore embedded commands, this provides a surface for indirect prompt injection.
[COMMAND_EXECUTION]: The skill executes external system binaries, including soffice, git, and pdftoppm, through subprocess.run(). This reliance on system-level command execution across various scripts increases the potential attack surface.

pptx