imaging-data-commons

Warn

Audited by Gen Agent Trust Hub on Apr 11, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides Python code that uses subprocess.run to execute shell commands for upgrading the idc-index package (pip3 install --upgrade --break-system-packages idc-index). This allows the skill to modify the agent's underlying execution environment at runtime.
  • [REMOTE_CODE_EXECUTION]: The instruction to install a package from a public registry and then immediately import it into the agent's process constitutes a remote code execution pattern.
  • [EXTERNAL_DOWNLOADS]: The skill manages the retrieval of medical imaging datasets from public AWS S3 and Google Cloud Storage buckets. These operations target well-known cloud services for legitimate scientific purposes but involve downloading large volumes of external data.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection due to the ingestion of metadata from external imaging and clinical databases.
  • Ingestion points: Data enters the agent's context through SQL queries via client.sql_query() and clinical data retrieval via client.get_clinical_table().
  • Boundary markers: The skill does not implement delimiters or explicit 'ignore instructions' warnings when processing database metadata.
  • Capability inventory: The environment includes high-privilege capabilities such as shell command execution (subprocess.run), opening web browsers (webbrowser.open), and file system writes.
  • Sanitization: There is no evidence of metadata sanitization or validation before the content is processed by the AI agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 11, 2026, 10:28 PM