tooluniverse-dataset-discovery
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (CSV, JSON, Excel, SAS, and Parquet files) retrieved from external URLs.
- Ingestion points: Data is fetched using
requests.get()and loaded viapandasloading functions as described in the loading cookbook inSKILL.md. - Boundary markers: No explicit delimiters or instructions are provided to the agent to treat external data as untrusted or to ignore potentially embedded instructions.
- Capability inventory: The skill possesses network read capabilities, file system write access, and the ability to execute arbitrary Python and Bash commands.
- Sanitization: There is no evidence of data sanitization, schema validation, or filtering for natural language instructions embedded within the datasets.
- [DYNAMIC_EXECUTION]: The skill provides structured code templates and explicitly instructs the agent to generate and run Python code at runtime to perform statistical analysis and data cleaning.
- [COMMAND_EXECUTION]: The instructions direct the agent to utilize shell environments (Bash) to execute the generated Python scripts, increasing the risk of command injection if parameters are not properly handled.
Audit Metadata