The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied .docx files, which constitutes a potential attack surface for indirect prompt injection if the agent were to interpret content within the XML as instructions. However, the skill's operations are confined to programmatic XML editing.
Ingestion points: Multiple scripts (e.g., fix_footnotes.py, check_crossrefs.py, docx_links.py) ingest and parse XML content from user-provided .docx archives using zipfile and lxml.
Boundary markers: The skill does not employ specific LLM boundary markers as it primarily functions as a standalone file-processing tool.
Capability inventory: The skill has the capability to write and overwrite files on the local filesystem and to execute internal Python scripts using subprocess.run.
Sanitization: The skill uses structured XML parsing via lxml and targeted regular expressions to isolate and modify specific document elements, which provides a layer of structural validation.
[COMMAND_EXECUTION]: The fix_footnotes.py script executes an internal command via subprocess.run to chain the create_crossrefs.py script. This is a standard and controlled execution of a script bundled within the same skill package.

docx-repair