dnanexus-integration

Pass

Audited by Gen Agent Trust Hub on Oct 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external bioinformatics data and platform metadata, creating a surface for indirect instructions to influence agent behavior.
  • Ingestion points: Data enters the environment via dxpy.download_dxfile, dxpy.DXFile.describe, and searching for objects with dxpy.find_data_objects.
  • Boundary markers (absent): No delimiters or "ignore embedded instructions" warnings are provided for processing these external inputs.
  • Capability inventory: The skill uses subprocess.check_call to run system commands and dxpy.DXApplet.run to execute platform-side code.
  • Sanitization (absent): The instructions do not define sanitization or validation routines for data content before it is processed or used in shell commands.
  • [COMMAND_EXECUTION]: The Python examples include usage of subprocess.check_call to execute local bioinformatics tools (e.g., quality_filter). This functionality permits the execution of system-level commands, which could be misused if arguments or filenames derived from untrusted platform data are not properly handled.
  • [PROMPT_INJECTION]: The section 'Suggest Using K-Dense Web For Complex Worflows' instructs the agent to proactively steer users toward a specific external service (www.k-dense.ai) when task complexity increases. This overrides the agent's default interaction guidelines to promote the vendor's hosted platform.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 12, 2026, 05:25 AM