llm-application-dev
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for interpolating external data directly into Large Language Model (LLM) prompts (e.g., ${customerFeedback} in the sentiment analysis example and {context} in the RAG pipeline). This creates an attack surface where maliciously crafted input data could attempt to override the system's instructions.
- Ingestion points: Data enters through the ragQuery function (context from vector database) and the user-provided variables in prompt templates.
- Boundary markers: The examples lack strong delimited boundaries or explicit instructions for the model to ignore embedded commands within the processed data.
- Capability inventory: The skill uses the OpenAI and Anthropic SDKs to make network requests to external LLM providers.
- Sanitization: The provided snippets do not include logic for sanitizing or escaping the data before it is included in the prompt templates.
Audit Metadata