The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill extracts natural language content from user-provided presentation files, which may contain instructions designed to manipulate the agent's behavior.\n
Ingestion points: Slide text is extracted in scripts/inventory.py and converted to markdown using the markitdown tool as described in SKILL.md.\n
Boundary markers: Absent. The skill's instructions do not specify the use of delimiters or provide warnings for the agent to disregard instructions embedded within presentation text.\n
Capability inventory: The skill possesses significant capabilities, including the ability to execute shell commands via subprocess.run (in pack.py, thumbnail.py, and redlining.py) and perform extensive file system modifications.\n
Sanitization: Extracted slide content is passed back to the agent's context without filtering or sanitization.\n- [COMMAND_EXECUTION]: The skill invokes external binaries to perform document conversion, validation, and comparison tasks.\n
Evidence: ooxml/scripts/pack.py and scripts/thumbnail.py call soffice (LibreOffice). scripts/thumbnail.py also utilizes pdftoppm for thumbnail generation. ooxml/scripts/validation/redlining.py uses git diff for comparing XML versions.\n
Context: These calls are implemented using list-based argument passing, which is safe from shell injection, and they are essential for the skill's primary functionality of presentation management.

pptx