The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains a hardcoded C source string (_SHIM_SOURCE) that is written to a temporary file and compiled into a shared object (lo_socket_shim.so) at runtime using gcc.
[PRIVILEGE_ESCALATION]: The skill uses LD_PRELOAD to inject the dynamically compiled C library into the soffice process. This technique is used to bypass system-level restrictions on AF_UNIX sockets in restricted or sandboxed environments, which constitutes a deliberate escalation of process capabilities beyond environment defaults.
[COMMAND_EXECUTION]: The skill utilizes several subprocess calls to manage document processing, including gcc for compilation, soffice for PDF conversion, and pdftoppm for image extraction.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: User-provided .pptx files are ingested into the agent context via the markitdown library as specified in SKILL.md.
Boundary markers: There are no boundary markers or instructions to ignore embedded commands within the extracted presentation text.
Capability inventory: The skill has significant capabilities including subprocess execution (gcc, soffice, pdftoppm) and file system write access (pack.py, add_slide.py).
Sanitization: No sanitization or filtering logic is present for the text extracted from slide decks.
[METADATA_POISONING]: The skill exhibits conflicting metadata; while the YAML frontmatter claims authorship by eigent-ai and a proprietary license, the LICENSE.txt file contains an Anthropic copyright and standard license terms, which may lead to confusion regarding the provenance and safety of the code.

pptx