The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes externally provided Word documents and their metadata, which creates a surface for indirect prompt injection.
Ingestion points: Document content is ingested via read_file, unzip, and pandoc conversion in SKILL.md.
Boundary markers: The skill includes a defensive guardrail (ASI01) that explicitly instructs the agent to treat extracted document text and comment threads as untrusted data.
Capability inventory: The skill possesses file system tools (write_file, edit_file) and shell execution capabilities.
Sanitization: Security guardrails instruct the agent to treat XML strictly as data and reject unverified active content such as macros (ASI04, ASI05).
[COMMAND_EXECUTION]: The skill instructions involve the use of standard command-line tools for document management.
Evidence: Instructions use pandoc, unzip, zip, and libreoffice for conversion and archive manipulation.
[DYNAMIC_EXECUTION]: The skill workflow includes the generation of temporary scripts for document creation.
Evidence: The "Suggested Process" section mentions authoring Node.js scripts using the docx library's Packer.toBuffer method.

docx