ai-native-development
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The calculate tool implementation in references/agentic-workflows.md uses the JavaScript eval() function to process mathematical expressions. This pattern allows for arbitrary code execution if the agent is manipulated into passing non-mathematical, malicious strings to the tool.
- [INDIRECT_PROMPT_INJECTION]: The skill implements Retrieval-Augmented Generation (RAG) and web search capabilities across multiple files, including templates/rag-pipeline-template.ts and templates/agent-workflow-template.ts. These features ingest untrusted data from external databases and the internet, which could contain hidden adversarial instructions. Mandatory Evidence Chain: 1. Ingestion points: retrieveDocuments in templates/rag-pipeline-template.ts and search_web in templates/agent-workflow-template.ts. 2. Boundary markers: System prompts instruct the agent to use only provided context. 3. Capability inventory: The agent can perform API calls to vector databases and LLM providers. 4. Sanitization: Input validation using Zod is discussed in references/function-calling.md but is not implemented in the base templates.
Audit Metadata