denario
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- External Downloads (HIGH): The skill promotes installation from untrusted sources, including the GitHub repository
AstroPilot-AI/Denarioand a Docker image (pablovd/denario:latest) from an individual contributor's account, which are not listed among trusted organizations. - Remote Code Execution (LOW): Documentation in
references/llm_configuration.mdsuggests installing the Google Cloud SDK usingcurl | bash. Although Google is a trusted organization, the piped execution method is inherently risky. - Command Execution (MEDIUM): The installation guide for LaTeX dependencies involves the use of
sudo, which requires elevated system privileges. - Dynamic Execution (MEDIUM): The skill utilizes AG2 and LangGraph to automatically generate and execute research methodologies and experiments, which involves running dynamically generated code at runtime.
- Indirect Prompt Injection (LOW): Untrusted research data and ideas ingested via
set_data_descriptionorset_ideacould contain malicious instructions that influence the code generated and executed during the research pipeline. Evidence: 1. Ingestion points:set_data_descriptionandset_ideainSKILL.md. 2. Boundary markers: Not specified. 3. Capability inventory: AI-driven computational experiment execution inget_results. 4. Sanitization: No sanitization methods are documented.
Recommendations
- AI detected serious security threats
Audit Metadata