The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/officecli_xml_common.py uses subprocess.run to execute the officecli command-line utility. This is the primary mechanism for the skill's intended document editing functionality. The implementation uses a list of arguments and does not invoke a shell (shell=True is absent), which mitigates common command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of JSON specification files and .docx documents, creating a potential surface for indirect injection.
Ingestion points: Command-line arguments pointing to document paths and JSON spec files (e.g., assets/specs/add_bookmarks.example.json) are used to drive the XML mutations.
Boundary markers: No explicit markers are used within the data files to separate instructions from content.
Capability inventory: The skill allows the agent to read and write .docx files through the officecli tool.
Sanitization: The raw_set function in scripts/officecli_xml_common.py includes a safety check that refuses empty XML payloads for 'replace' actions to prevent accidental document corruption. The scripts also perform basic validation on the JSON input, such as ensuring unique bookmark IDs.
[SAFE]: The skill demonstrates security best practices for its domain, including clear warnings about the destructive potential of raw XML edits and detailed documentation in the references/ directory to ensure the agent follows valid OOXML (Office Open XML) structures. It also uses the uv tool for reproducible execution environments.

sdo-docx-operations