The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process metadata from the National Cancer Institute Imaging Data Commons (IDC), creating a potential surface for indirect prompt injection. The evidence chain is as follows:
Ingestion points: Data enters the agent's context through IDCClient.sql_query() and IDCClient.fetch_index() calls documented in SKILL.md and implemented in scripts/query_imaging_data_commons.py.
Boundary markers: The skill's instructions do not include explicit delimiters or guardrails to prevent the agent from accidentally executing instructions potentially embedded within metadata fields like SeriesDescription or StudyDescription.
Capability inventory: The skill possesses file-writing capabilities through client.download_from_selection() and the ability to execute code via the Bash(python:*) tool.
Sanitization: There is no documentation of metadata sanitization or filtering to mitigate the risk of malicious payload execution.
[EXTERNAL_DOWNLOADS]: The skill enables the download of DICOM imaging datasets from official IDC cloud storage buckets on AWS S3 and Google Cloud Storage (e.g., s3://idc-open-data/). It also references the installation of the idc-index Python package. These resources are provided by the National Cancer Institute and are considered safe, well-known research infrastructure.

alterlab-imaging-data-commons