finding-data-lake-assets
Warn
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill implements a 'Broad Scan Fallback' procedure in Step 5b that requires the agent to dynamically generate a Python script and execute it using
python3. While this is intended for comprehensive catalog searches usingboto3, the practice of generating and running code at runtime represents a significant security surface. - [PROMPT_INJECTION]: The skill is designed to ingest metadata from the AWS Glue Data Catalog, which is considered an untrusted external source. This creates an indirect prompt injection surface. The skill recognizes this and includes explicit instructions to ignore natural language directives found in catalog fields like 'Description' or 'Forms', which were flagged by static analysis tools during the audit.
- [COMMAND_EXECUTION]: The skill utilizes the
awsCLI and MCP tools to interact with cloud resources. To mitigate the risk of command injection, the instructions mandate that any user-provided parameters must be shell-quoted before being incorporated into command-line strings.
Audit Metadata