giskard-llm
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the
giskardlibrary from PyPI using the commandpip install "giskard[llm]>2,<3". Giskard is a widely recognized open-source framework for testing and evaluating AI models. References are made to the official Giskard-AI GitHub repository and documentation site. - [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow for generating and testing adversarial prompts against an LLM.
- Ingestion points: The
model_predictfunction inSKILL.md(Step 2) accepts a DataFrame containing user-provided questions. - Boundary markers: None explicitly defined in the boilerplate code, as the tool is designed to generate the injection inputs for testing purposes.
- Capability inventory: The skill generates an HTML report using
scan_results.to_html(). No high-privilege capabilities or dangerous system access patterns are present. - Sanitization: The tool is intended to identify vulnerabilities in how the target model handles unsanitized inputs, serving as a security testing utility.
Audit Metadata