haystack

Pass

Audited by Gen Agent Trust Hub on Mar 11, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The documentation and scripts recommend installing the haystack-ai and sentence-transformers libraries via standard package managers. These are well-known, legitimate open-source libraries for building AI search systems.
  • [PROMPT_INJECTION]: The skill demonstrates the construction of Retrieval-Augmented Generation (RAG) pipelines in scripts/basic_rag_pipeline.py. This introduces a standard indirect prompt injection surface as user queries and retrieved documents are interpolated into LLM prompts.
  • Ingestion points: User input (query) and retrieved data (documents) in scripts/basic_rag_pipeline.py and references/components.md.
  • Boundary markers: The templates use standard text headers like "Context:" and "Question:" to delimit content.
  • Capability inventory: The pipeline has the capability to generate text via the OpenAIChatGenerator based on the provided context.
  • Sanitization: The example code does not implement explicit sanitization of retrieved content, which is typical for framework-level documentation but represents a surface to be aware of during implementation.
  • [CREDENTIALS_UNSAFE]: The skill correctly implements secret management by using Haystack's Secret.from_env_var utility in scripts/basic_rag_pipeline.py, ensuring that sensitive API keys are not hardcoded in the source code.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 11, 2026, 05:54 PM