The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts convert.py and import_to_wps.py use subprocess.run to interface with external tools like pandoc, pdfimages, and wpsnote-cli. These calls use list-based arguments to avoid shell injection and are integral to the document conversion and import process.
[EXTERNAL_DOWNLOADS]: The documentation references several well-known and trusted third-party libraries (e.g., beautifulsoup4, pdfplumber, pandas) and system utilities (pandoc, tesseract) required for document parsing. These are standard tools for the described tasks and are sourced from official repositories.
[DATA_EXFILTRATION]: The skill reads local user files (Markdown, PDF, DOCX, etc.) to convert and upload them to WPS Notes. This data flow is the core intended functionality of the document importer and is directed to the user's associated WPS account via a vendor-provided CLI tool.
[SAFE]: Analysis of the Python scripts confirms that file access and network interactions via the CLI tool are consistent with the skill's purpose. The scripts use standard XML escaping and temporary directory management. No evidence of malicious prompt injection, obfuscated code, or unauthorized persistence was found.
[SAFE]: Regarding indirect prompt injection surfaces: 1. Ingestion points: Local document files (.md, .html, .pdf, etc.) scanned from user-specified directories. 2. Boundary markers: Not used for processed content. 3. Capability inventory: Subprocess execution, file system access, and WPS API interaction. 4. Sanitization: The skill applies XML escaping to extracted text to maintain format integrity, but does not specifically filter for LLM-targeting injection patterns in the imported content. This is considered a safe surface given the context of a personal note-importing utility.

doc-importer