The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py generates C source code from a hardcoded string at runtime and compiles it using gcc into a shared object library. This allow the creation of arbitrary native code.
[COMMAND_EXECUTION]: The script scripts/office/soffice.py utilizes the LD_PRELOAD environment variable to inject the compiled shared library into the soffice (LibreOffice) process execution. This mechanism is used to intercept system calls and modify process behavior at a low level.
[COMMAND_EXECUTION]: Multiple scripts (scripts/pptx_render.py, scripts/thumbnail.py, scripts/office/soffice.py, scripts/office/validators/redlining.py) use subprocess.run to execute system tools like libreoffice, pdftoppm, gcc, and git for rendering, file conversion, and comparison tasks.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from PowerPoint files.
Ingestion points: Untrusted text and media are read in scripts/pptx_text.py and scripts/pptx_modify.py.
Boundary markers: Absent when text is presented to the agent for summarization or inspection.
Capability inventory: Use of subprocess.run in scripts/pptx_render.py and scripts/thumbnail.py provides an exploitation path if the agent is manipulated.
Sanitization: The skill uses defusedxml for secure XML parsing, which mitigates XML-based attacks but not logical prompt injection.

pptx