The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py includes a hardcoded C source code block that is written to a temporary file and compiled at runtime using gcc. The resulting shared object is then loaded into the soffice (LibreOffice) process via the LD_PRELOAD environment variable. While this is intended to bypass AF_UNIX socket restrictions in sandboxed environments, runtime compilation and library injection into subprocesses represent high-risk execution patterns.
[METADATA_POISONING]: The skill's documentation and frontmatter metadata claim it is created and maintained by "Anthropic, PBC", providing official Anthropic source URLs. However, the associated context identifies the author as "k-dense-ai". This discrepancy in authorship is deceptive and may lead users to misjudge the skill's trust level.
[COMMAND_EXECUTION]: The skill frequently invokes external binaries through the subprocess module. These include soffice for PDF conversion, pdftoppm for image generation, git for diffing content, and gcc for the runtime compilation described above. These executions are integrated into the primary workflow of presentation analysis and structural duplication.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied .pptx files as its primary input.
Ingestion points: Presentation content is ingested in scripts/thumbnail.py (line 120) and scripts/office/validate.py during parsing.
Boundary markers: The instructions do not define boundary markers or

pptx