imaging-data-commons
Warn
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The
SKILL.mdfile contains an embedded Python script that usessubprocess.runto executepip3 installwith the--break-system-packagesflag. This allows the agent to execute arbitrary shell commands that modify the system's global Python environment, bypassing standard safety constraints. - [REMOTE_CODE_EXECUTION]: The skill automates the installation and upgrading of the
idc-indexpackage from a remote registry at runtime. Dynamic installation of third-party software creates a significant attack surface for supply chain exploitation. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from various databases, including slide microscopy annotations (
AnnotationGroupLabel) and clinical records. These fields are processed by the agent without explicit boundary markers or sanitization, potentially allowing malicious data to influence agent behavior. - [EXTERNAL_DOWNLOADS]: Multiple components of the skill, including guides for BigQuery, Cloud Storage, and Parquet access, involve fetching data from external cloud providers (GCS/AWS). While these appear to be related to the service, the skill establishes multiple network connections to fetch executable or structured data.
Audit Metadata