ai-ml-integration
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements several Retrieval-Augmented Generation (RAG) patterns in
SKILL.mdthat interpolate untrusted external data and user-provided queries into prompts without sufficient sanitization. - Ingestion points: Data is ingested via
PDFLoaderandPineconeStore.similaritySearchinSKILL.md(lines 315-325, 405-420), as well as through user queries in theRAGPipeline.queryandAdvancedRAG.querymethods. - Boundary markers: The implementations use simple newline-separated text delimiters (e.g.,
Context: ${context}\n\nQuestion: ${question}) which provide minimal protection against adversarial context injection. - Capability inventory: The skill utilizes LLM API capabilities (
openai,anthropic,cohere) to process the aggregated and potentially poisoned context. - Sanitization: There is no evidence of input validation, escaping, or filtering of the retrieved document content before it is processed by the language model.
Audit Metadata