The Agent Skills Directory

[SAFE]: Secure XML handling is integrated throughout the skill. Scripts such as extract_story_text.py, idml_document.py, and the utilities.py module utilize defusedxml for all XML parsing operations. This effectively neutralizes risks associated with XML External Entity (XXE) and billion laughs attacks that could be present in malicious IDML files.
[SAFE]: Path traversal protection in archive processing. The unpack_idml.py script includes a robust safety check (_is_safe_path) that validates archive members against the target directory, preventing malicious IDML files from writing data to sensitive locations outside the intended workspace.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of Adobe InDesign (IDML) packages and their constituent XML files, creating a potential surface for indirect prompt injection. However, the risk is mitigated by the skill's technical safeguards and its focused application on structured layout and text editing.
Ingestion points: Stories, spreads, and resource XML files processed within the scripts/ directory.
Boundary markers: Absent; the skill processes data within fixed XML schemas.
Capability inventory: Subprocess script execution (via runpy in smoke_test.py), file system read/write, and XML node manipulation.
Sanitization: Comprehensive XML sanitization via defusedxml and filename sanitization in unpack_idml.py.

idml