The Agent Skills Directory

[SAFE]: The skill is a well-structured tool for paper translation. It uses a Python script (scripts/pages.py) to manage local task data, render source pages, and assemble final PDFs from generated images. It does not perform network operations or access sensitive system paths.
[COMMAND_EXECUTION]: A deterministic detector flagged subprocess.run in tests/test_pages.py. Analysis confirms this is a safe implementation used exclusively within the test suite to execute the skill's own script via sys.executable for functional verification.
[DATA_EXFILTRATION]: No data exfiltration risks were found. The script scripts/pages.py employs a path-normalization function (asset) that enforces a strict workspace boundary, preventing directory traversal attacks. All file operations are local and verified against SHA256 hashes.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external PDF files. The implementation mitigates this by rendering the content into a static PNG before it is passed to the image model. The skill also enforces a mandatory review step where the agent checks the output for content or layout anomalies.
[DYNAMIC_EXECUTION]: The script uses standard libraries for its functionality. The use of importlib in the test suite is for loading the script module during unit testing, which is a standard and benign development practice.

nature-paper-trans