The Agent Skills Directory

[DYNAMIC_EXECUTION]: Runtime compilation and library injection detected in scripts/office/soffice.py.
The script defines a C source string (_SHIM_SOURCE) designed to shim UNIX socket calls.
It dynamically writes this source to a temporary file and invokes gcc to compile a shared object (.so) file.
It then uses the LD_PRELOAD environment variable to inject this library into the soffice process at runtime.
This mechanism is intended to allow LibreOffice to operate in environments where standard socket operations are restricted, which is essential for the skill's rendering functionality.
[INDIRECT_PROMPT_INJECTION]: Surface area for untrusted data processing.
Ingestion points: The skill processes .pptx files provided as user input (e.g., in scripts/office/unpack.py and scripts/thumbnail.py).
Boundary markers: Absent. There are no explicit delimiters or instructions to the model to ignore potentially malicious text embedded within slide XML.
Capability inventory: The skill has the capability to execute arbitrary subprocesses (soffice, pdftoppm, git, gcc) and perform extensive file system operations.
Sanitization: The skill utilizes defusedxml for XML parsing, which provides strong protection against XML External Entity (XXE) attacks, but does not sanitize the logical content of the slides.
[COMMAND_EXECUTION]: The skill relies heavily on external binary execution.
Commands like soffice, pdftoppm, and git are executed via subprocess.run to handle file conversion, image generation, and version comparisons. This increases the risk if vulnerabilities exist in the underlying system tools when processing malformed binary inputs.

pptx