omh-data-analysis
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external datasets, providing an entry point for untrusted content that could attempt to influence agent behavior.
- Ingestion points: The skill triggers on analysis requests for datasets, tables, CSVs, JSON, and system logs.
- Boundary markers: The instructions explicitly mandate the use of 'not-evidence boundaries' and 'prepared-vs-observed boundaries' to isolate preparation logic from data-derived findings.
- Capability inventory: The skill utilizes data processing tools (pandas, SQL) and command outputs to perform calculations and schema checks.
- Sanitization: The workflow requires the agent to print data characteristics (shape, dtypes, null counts, duplicate rows, and min/max values) before making any claims, ensuring claims are rooted in verifiable observations.
- [COMMAND_EXECUTION]: The skill relies on the execution of data analysis libraries and command-line tools to provide accurate results.
- Evidence: The instructions specify that 'every number in the answer comes from a command output you can name' and that data should be 'loaded with code (pandas, SQL or a spreadsheet tool).'
Audit Metadata