structured-output
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run commands with a --help flag during steps 2, 3, and 4 to verify tool availability. These are limited, non-exploitative checks for ensuring the formatting tools are present and accessible.
- [DYNAMIC_EXECUTION]: The instructions reference the execution of a specific local script, 'plugins/imbue/scripts/citation_verifier.py', to mechanically verify verbatim text anchors in findings. This script is part of the internal 'imbue' framework and is called via a static file path.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external analysis results and source code snippets to generate structured reports. While this establishes a data ingestion surface, the risk is low as the skill's instructions focus on formatting and do not execute the processed content.
- Ingestion points: Analysis findings and verbatim source text anchors processed during output generation (SKILL.md).
- Boundary markers: None explicitly defined in the template instructions for separating finding descriptions from external content.
- Capability inventory: Execution of local citation verifier scripts and tool availability checks using --help flags (SKILL.md).
- Sanitization: No sanitization of the input text or verbatim anchors is specified before interpolation into report templates.
Audit Metadata