AgentDB Vector Search

Pass

Audited by Gen Agent Trust Hub on Apr 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The instructions guide the agent or user to execute various shell commands using npx agentdb@latest for initializing the database, querying vectors, and managing exports/imports.
  • [EXTERNAL_DOWNLOADS]: The skill relies on npx agentdb@latest and agentic-flow, which involve fetching and executing packages from the npm registry at runtime.
  • [PROMPT_INJECTION]: The provided RAG (Retrieval Augmented Generation) code example in the 'Advanced Usage' section contains a vulnerability to indirect prompt injection by directly interpolating retrieved context into an LLM prompt.
  • Ingestion points: Untrusted data is retrieved from the vector database using db.searchSimilar in SKILL.md.
  • Boundary markers: The example lacks delimiters (e.g., XML tags or triple quotes) and instructions for the LLM to ignore potentially malicious commands within the retrieved text.
  • Capability inventory: The skill facilitates passing retrieved data directly to an LLM via the llm.generate method.
  • Sanitization: No input validation or sanitization logic is demonstrated or recommended for the retrieved document content before it is processed by the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 2, 2026, 07:18 AM
Security Audit — agent-trust-hub — AgentDB Vector Search