The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes text extracted from external PDF documents, which creates a surface for indirect prompt injection attacks where malicious instructions could be embedded in the PDF content.
Ingestion points: scripts/paper_translator.py extracts text from user-provided PDF files using pdftotext or the pypdf library.
Boundary markers: The script wraps the extracted text in markdown code blocks within the translation prompt, but it lacks explicit instructions for the model to ignore potentially adversarial content inside the text.
Capability inventory: The skill can read local files, write to the filesystem (for PDFs and cache), execute subprocesses (pdftotext), and perform network operations to translation APIs.
Sanitization: Basic whitespace and control character cleaning is performed on the extracted text, but there is no logic to detect or neutralize prompt injection attempts within the content.
[COMMAND_EXECUTION]: The script uses an external utility to perform text extraction from PDF documents.
Evidence: scripts/paper_translator.py executes the pdftotext utility using subprocess.run. The implementation uses a list of arguments rather than a shell string, which is a secure method that prevents shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill documentation refers to the installation of a third-party tool for enhanced translation capabilities.
Evidence: SKILL.md suggests using uv tool install to install pdf2zh, which is a legitimate and well-known utility for preserving PDF layout during translation.

lov-paper-translator