ngs-analysis-router
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [Indirect Prompt Injection Surface]: The skill processes untrusted biological metadata files (e.g.,
SampleSheet.csv,RunInfo.xml) to determine pipeline parameters. This creates a surface where malicious instructions could be embedded in these files to influence the agent's behavior. - Ingestion points: The agent is instructed to read
SampleSheet.csv,RunInfo.xml,RunParameters.xml, and various sequencing metadata files. - Boundary markers: There are no explicit instructions to the agent to treat content within these files as data rather than instructions, increasing the risk of indirect injection.
- Capability inventory: The agent can execute local Python scripts (
ngs_preflight.py,ngs_reference_manager.py) and generate shell scripts (install_commands.sh) which can be executed if the user provides a specific environment variable. - Sanitization: The instructions do not specify any validation or sanitization steps for the data extracted from these input files before using them in shell command parameters.
- [Dynamic Script Generation and Execution]: The skill generates a shell script (
install_commands.sh) for pipeline installation. While execution is protected by an environment variable (NGS_RUN_INSTALL_COMMANDS=1), the content of the script is derived from the agent's interpretation of the environment and input files. - [Potential Data Exposure]: The skill handles sensitive genomic data and references cloud analysis platforms (e.g., BaseSpace, Terra, DNAnexus). While it explicitly requires the agent to ask for permission before cloud uploads, the capability to move sensitive data to external services is a notable security consideration.
Audit Metadata