knowledge-graph-builder

Pass

Audited by Gen Agent Trust Hub on Feb 24, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill describes patterns for ingesting unstructured text to build and query knowledge graphs, which represents a standard indirect prompt injection surface characteristic of retrieval-augmented generation (RAG) systems.
  • Ingestion points: Untrusted text is processed in references/entity-extraction.md and references/ai-integration.md during entity extraction and subgraph retrieval.
  • Boundary markers: The LLM prompts in references/ai-integration.md use basic headers (e.g., 'User query:') but do not demonstrate the use of robust delimiters or explicit instructions to ignore embedded commands.
  • Capability inventory: The architectural patterns provided do not include scripts with dangerous capabilities like arbitrary command execution or unauthorized network access.
  • Sanitization: The provided examples focus on architecture and do not include specific input validation or sanitization logic.
  • [EXTERNAL_DOWNLOADS]: The documentation references integration with several well-known and trusted technology services for graph databases, vector storage, and machine learning.
  • Evidence: References include Neo4j, Amazon Neptune, ArangoDB, TigerGraph, Pinecone, OpenAI, spaCy, and Hugging Face across files such as references/database-selection.md and references/hybrid-architecture.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 24, 2026, 08:36 PM