The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes several Python scripts for DICOM analysis and transformation. These scripts use local-only command-line interfaces (CLIs) that do not spawn arbitrary shell commands, use subprocess, or interact with the system shell in an unsafe manner. All operations are confined to standard library file I/O.
[EXTERNAL_DOWNLOADS]: The installation instructions reference well-known and trusted Python packages such as pydicom, numpy, and Pillow, as well as specific transfer syntax plugins from the PyPI registry. The analysis confirms no scripts perform runtime downloads or piping of remote content to execution engines.
[DATA_EXFILTRATION]: The skill is designed to handle Protected Health Information (PHI) in DICOM files. It mitigates exfiltration risks by explicitly disabling network access across all scripts. I/O is restricted to a user-defined local root directory, and the skill implements strict path validation that rejects parent directory traversal (..), symlinks, and non-local URI prefixes (e.g., http:, s3:).
[CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or private secrets were detected. The skill provides a utility to generate local deterministic keys for pseudonymization, which are stored with owner-only (0600) file permissions on supported systems, aligning with secure secret management practices.
[DYNAMIC_EXECUTION]: The skill performs runtime inspection of installed DICOM codecs and plugins using importlib.metadata. This is a standard technical inventory task and does not involve the use of eval(), exec(), or unsafe deserialization of untrusted data.

pydicom