timeseries-modeling
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external CSV files to generate a diagnostics summary that the agent then uses. This creates a potential attack surface where malicious instructions could be embedded within data headers or content to influence the agent's reasoning.
- Ingestion points: The
scripts/timeseries_modeling_diagnostics.pyscript reads user-providedinput.csvfiles. - Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions embedded in the data.
- Capability inventory: The skill utilizes local script execution and file system read/write operations.
- Sanitization: While the script performs type conversion for numeric data, it does not sanitize metadata such as column names, which are included in the final artifact ingested by the agent.
Audit Metadata