The Agent Skills Directory

[METADATA_POISONING]: The skill's license file (LICENSE.txt) and instructions contain copyright claims and descriptions identifying 'Anthropic, PBC' as the creator, which contradicts the provided author context of 'foraefactory'. This discrepancy is deceptive and could lead users to misjudge the skill's origin and safety.
[INDIRECT_PROMPT_INJECTION]: The skill extracts and reads content from external .docx files into the agent's context, creating an attack surface for indirect prompt injection.
Ingestion points: XML files are extracted from user-supplied documents using ooxml/scripts/unpack.py and read by the agent using the scripts/document.py library.
Boundary markers: Absent. The instructions do not define delimiters or specific warnings for the agent to ignore instructions embedded within processed documents.
Capability inventory: The skill includes the ability to execute system commands via subprocess.run (in ooxml/scripts/pack.py and ooxml/scripts/validation/redlining.py) and perform extensive file system modifications.
Sanitization: While defusedxml is used to prevent XML-based attacks like XXE, the text content extracted from documents is not sanitized or filtered for prompt instructions.
[COMMAND_EXECUTION]: The skill uses the subprocess module to execute external utility commands for its workflow.
File: ooxml/scripts/pack.py. Evidence: Executes 'soffice' in headless mode to validate document integrity through conversion.
File: ooxml/scripts/validation/redlining.py. Evidence: Executes 'git diff' to perform character-level comparison between original and modified documents.
[EXTERNAL_DOWNLOADS]: SKILL.md recommends the installation of several well-known external dependencies via system package managers (apt) and NPM, including pandoc, libreoffice, and the docx library.

docx