The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains unit tests in tests/test_evidence_tools.py and tests/test_topic_tools.py that utilize subprocess.run to execute and verify the internal utility scripts. This usage is restricted to the skill's own development environment for testing purposes and does not introduce risk during normal operation.
[DATA_EXPOSURE]: File system operations in scripts/evidence_tools.py and scripts/topic_tools.py are securely implemented using pathlib with explicit boundary checks (is_relative_to(root)). This design prevents directory traversal attacks and ensures that the skill only interacts with files within the user-defined project directory.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external scientific literature (PDFs and Markdown), which serves as a potential vector for indirect prompt injection, it mitigates this risk through a structured 'Evidence Contract'. This system uses SHA-256 hashes to verify file integrity and requires explicit 'locators' (page/figure references) for every claim, ensuring that the agent remains grounded in verified source material.

sanmu-paper-reader