llm-application-dev
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill demonstrates RAG (Retrieval-Augmented Generation) and prompt engineering patterns that create a surface for indirect prompt injection.
- Ingestion points: The
ragQueryfunction and prompt templates inSKILL.mdtake untrusted input from variables likequestion(user input) andcontext(retrieved from a vector database). - Boundary markers: The code snippets lack robust delimiters (such as XML tags or unique string markers) or instructions to the model to ignore embedded instructions within the interpolated variables.
- Capability inventory: The skill includes integration patterns for the OpenAI and Anthropic SDKs, which facilitate active interaction with LLMs.
- Sanitization: No sanitization, escaping, or validation logic is present in the provided examples to mitigate the risk of a user or a retrieved document overriding the application's intended logic.
Audit Metadata