The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes instructions to verify and upgrade the idc-index Python package via subprocess.run. This is a legitimate environment setup step to ensure compatibility with the current IDC data version (v23).
[EXTERNAL_DOWNLOADS]: Fetches medical imaging data and metadata from public cloud storage buckets (AWS S3 and Google Cloud Storage) and BigQuery datasets maintained by the NCI Imaging Data Commons. These are well-known scientific resources.
[PROMPT_INJECTION]: The skill processes structured metadata from the IDC database and manifest files. While this represents an indirect prompt injection attack surface, the risk is mitigated by the trusted source of the data and the skill's focus on scientific research.
Ingestion points: SQL query results from client.sql_query() and manifest files (.txt, .csv) processed by the idc command-line tool.
Boundary markers: None explicitly defined; data is typically processed as structured Pandas DataFrames.
Capability inventory: Includes file system access (downloading DICOM files), network access (cloud storage), and subprocess execution (standard package installation).
Sanitization: Relies on the idc-index library's handling of metadata and the public nature of the National Cancer Institute repository.

imaging-data-commons