AgentDB Vector Search
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (MEDIUM): The skill instructions frequently use
npx agentdb@latestfor database operations, MCP setup, and benchmarking. - Evidence: Commands such as
npx agentdb@latest init ./vectors.dbandnpx agentdb@latest mcpare used throughout the skill. - Risk: Using
@latestwithnpxdownloads and executes code from the npm registry without version pinning. Since the package author/organization (ruvnet) is not on the trusted list, this poses a supply chain risk if the package were compromised. - PROMPT_INJECTION (LOW): The skill is explicitly designed for Retrieval-Augmented Generation (RAG), creating a surface for indirect prompt injection.
- Ingestion points: Untrusted data is ingested into the database via
storeWithEmbeddingor CLIimportcommands (found inSKILL.md). - Boundary markers: The
ragQueryexample shows context being directly interpolated into a prompt string (Context: ${context.map(c => c.text).join('\n')}) without any delimiters or 'ignore' instructions. - Capability inventory: The skill enables the agent to fetch stored documents and pass them to
llm.generate()for response synthesis. - Sanitization: No sanitization or validation of the retrieved content is performed before it is presented to the LLM.
Audit Metadata